diagnostics
Article
Can Artificial Intelligence “Hold” a Dermoscope?—The
Evaluation of an Artificial Intelligence Chatbot to Translate
the Dermoscopic Language
Emmanouil Karampinis 1, * , Olga Toli 2 , Konstantina-Eirini Georgopoulou 3 , Elli Kampra 1 ,
Christina Spyridonidou 4 , Angeliki-Victoria Roussaki Schulze 1 and Efterpi Zafiriou 1, *
Citation: Karampinis, E.; Toli, O.;
Georgopoulou, K.-E.; Kampra, E.;
Spyridonidou, C.; Roussaki Schulze,
A.-V.; Zafiriou, E. Can Artificial
Intelligence “Hold” a Dermoscope?—
The Evaluation of an Artificial
Intelligence Chatbot to Translate the
Dermoscopic Language. Diagnostics
2024, 14, 1165. https://doi.org/
10.3390/diagnostics14111165
Academic Editor: Dechang Chen
Received: 20 April 2024
Revised: 24 May 2024
Accepted: 30 May 2024
Published: 31 May 2024
Copyright: © 2024 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Diagnostics 2024, 14, 1165. https://doi.org/10.3390/diagnostics14111165
1 Department of Dermatology, Faculty of Medicine, School of Health Sciences, University General Hospital of
Larissa, University of Thessaly, 41110 Larissa, Greece; kampraelli@gmail.com (E.K.);
roussaki@otenet.gr (A.-V.R.S.)
2 Department of Dermatology, Oncoderm Center One Day Clinic, 45332 Ioannina, Greece;
olgatolimail@gmail.com
3 Department of Dermatology, General Hospital of West Attica “Agia Varvara”, 12351 Athens, Greece;
koneirgeo@gmail.com
4 Department of Dermatology, Athens Naval Hospital, 11521 Athens, Greece
* Correspondence: ekarampinis@uth.gr (E.K.); zafevi@o365.uth.gr (E.Z.)
Abstract: This survey represents the first endeavor to assess the clarity of the dermoscopic language
by a chatbot, unveiling insights into the interplay between dermatologists and AI systems within the
complexity of the dermoscopic language. Given the complex, descriptive, and metaphorical aspects
of the dermoscopic language, subjective interpretations often emerge. The survey evaluated the
completeness and diagnostic efficacy of chatbot-generated reports, focusing on their role in facilitating
accurate diagnoses and educational opportunities for novice dermatologists. A total of 30 participants
were presented with hypothetical dermoscopic descriptions of skin lesions, including dermoscopic
descriptions of skin cancers such as BCC, SCC, and melanoma, skin cancer mimickers such as actinic
and seborrheic keratosis, dermatofibroma, and atypical nevus, and inflammatory dermatosis such
as psoriasis and alopecia areata. Each description was accompanied by specific clinical information,
and the participants were tasked with assessing the differential diagnosis list generated by the AI
chatbot in its initial response. In each scenario, the chatbot generated an extensive list of potential
differential diagnoses, exhibiting lower performance in cases of SCC and inflammatory dermatoses,
albeit without statistical significance, suggesting that the participants were equally satisfied with
the responses provided. Scores decreased notably when practical descriptions of dermoscopic signs
were provided. Answers to BCC scenario scores in the diagnosis category (2.9 ± 0.4) were higher
than those with SCC (2.6 ± 0.66, p = 0.005) and inflammatory dermatoses (2.6 ± 0.67, p = 0). Similarly,
in the teaching tool usefulness category, BCC-based chatbot differential diagnosis received higher
scores (2.9 ± 0.4) compared to SCC (2.6 ± 0.67, p = 0.001) and inflammatory dermatoses (2.4 ± 0.81,
p = 0). The abovementioned results underscore dermatologists’ familiarity with BCC dermoscopic
images while highlighting the challenges associated with interpreting rigorous dermoscopic images.
Moreover, by incorporating patient characteristics such as age, phototype, or immune state, the
differential diagnosis list in each case was customized to include lesion types appropriate for each
category, illustrating the AI’s flexibility in evaluating diagnoses and highlighting its value as a
resource for dermatologists.
Keywords: artificial intelligence; chatbots; dermoscopy; differential diagnosis
1. Introduction
Artificial intelligence (AI) is increasingly becoming popular in the section of dermatol-
ogy and has become a helping hand to dermatologists for skin cancer diagnosis as well
https://www.mdpi.com/journal/diagnostics
Diagnostics 2024, 14, 1165 2 of 21

as difficult cases. Currently, studies report high accuracy performance, even exceeding
dermatologists for the diagnosis of skin lesions [1]. The use of artificial intelligence in der-
matology is mainly image-based and is constructed by the deep learning (DL) methodology,
which makes connections between inputs such as images and outputs such as diagnoses
like BCC or melanoma [2]. As a result, a complex mapping is created with links that bond
image patterns and prediction of diagnosis. Deep learning is usually based on self-learning
and training through image databases that contain macroscopic or dermatoscopic images
of skin cancer subtypes [1,3]. The convolutional neural networks (CNNs), the leading DL
algorithm for image analysis, have reported a high performance rate in melanoma diagno-
sis [4]. Additionally, due to the unpredictable nature of inflammatory dermatoses [5], AI
technology has been able to predict the flares of skin diseases such as psoriasis and atopic
dermatosis [6].
Apart from the image-based and flare-up-predicting AI, which is the hallmark of
this type of technology in dermatology, chatbots are emerging as a new addition to the
dermatological domain. They offer a convenient and practical tool that is accessible not
only to clinicians at all levels of healthcare but also to patients. AI chatbots represent a novel
type of artificial intelligence that employs natural language comprehension techniques in
conversations with users when they pose a question. The methodology used by chatbots is
also based on neural networks that are trained on extensive text datasets [7]. Dermatology
is a domain that includes complicated terminology, ranging from complicated dermatosis
presentations, including primary and secondary cutaneous lesions, to the dermoscopic
language and dermatopathological reports. Therefore, how efficiently can a text-based
AI program such as a chatbot deal with questions regarding this “cutaneous” aspect
of medicine?
OpenAI’s ChatGPT is a program built by the chatbot model technology. The advanced
search of the combination of (ChatGPT) AND (medicine) results in more than 1500 studies
over a period of only 2 years. The chatbot operates on GPT-3.5 or GPT-4 (if available through
subscription). GPT-4, the most recent model released by OpenAI, boasts advancements over
its predecessor (ChatGPT-3.5). It is reported to excel in tackling more complex problems
with greater accuracy, attributed to enhanced reasoning capabilities and a significantly
expanded repository of learned information. However, GPT-3.5 is more popular due to the
lack of a need for a subscription [8,9].
Chatbots including GPT-3.5 and GPT-4 have been frequently used in dermatology,
usually for teaching purposes and patient confrontation. Most of the studies aim to present
complicated medical issues in a simplified language for the patients to understand or even
lead to self-diagnosis. ChatGPT can be helpful in aiding healthcare providers in formulating
differential diagnoses and therefore aid practitioners, and, more frequently, new-certified
clinicians, to formulate their approach to the patients. This approach is also expanded in
dermatology [9]. The differential diagnoses generated are also in question regarding their
completeness and contribution to diagnosis. ChatGPT can become virtual tutors and be
tailored to each student’s unique learning preferences and aptitudes. Additionally, ChatGPT
can generate dermatological scenarios and support students in their academic endeavors by
addressing their queries and crafting summaries about specific dermatological themes [10].
Chatbots have been utilized and investigated across various domains of dermatology,
as outlined in Box 1, encompassing tasks such as skin cancer detection, the monitoring of
inflammatory dermatoses, and applications in aesthetic medicine. Box 1 includes studies
published until 28 March 2024 on Pubmed regarding the use of chatbots in dermatology.
Specifically, patients often inquire about symptoms or the appearance of lesions they are
experiencing prior to consulting a physician, seeking to determine if they could signify ma-
lignancies or systemic illnesses [9]. If the responses from the chatbot suggest the possibility
of cancer or serious conditions, patients typically seek medical attention [10]. Additionally,
patients frequently seek further information about their condition, including its etiology,
adherence to treatment, whether it is contagious, and how it may impact their daily lives.
Diagnostics 2024, 14, 1165 3 of 21

Also, disease-specific counseling is considered as virtual healthcare assistants available at

any time, such as for hidradenitis suppurativa [11]).

Box 1. Box presenting the use of chatbots in different aspects of Dermatology practice.

Use of chatbots in dermatology

- Skin cancers
ChatGPT generation of medical information in response to 25 clinical questions about NMSC (non-
melanoma skin cancer): use of language learning models and zero-shot chain of thought (“Let’s
think step by step!” strategy) [12].
ChatGPT4 reproduced dermatohistopathological reports in a patient-friendly language effec-
tively [13].
- Inflammatory Dermatosis
Psoriasis
Investigation of the abilities of ChatGPT to compare the different systemic therapeutic interventions
for moderate-to-severe psoriasis [14].
Atopic Dermatitis
By comparing ChatGPT answers regarding atopic dermatitis and acne, precision and inclusiveness
were diminished for inquiries regarding acne treatment compared to atopic dermatitis (AD), occa-
sionally omitting details on treatment efficacy, lacking clarity on treatment outcomes, and neglecting
to offer tailored recommendations suitable for the patient’s age and individual circumstances [15].
Rosacea
ChatGPT 3.5 demonstrates remarkable reliability and practical relevance when addressing common
inquiries from patients regarding rosacea, as evidenced by the analysis of 20 questions–answers [16].
Hidradenitis
A medical hidradenitis chatbot can provide high-quality information within seconds [11].
Allergy
The effective utilization of a chatbot to streamline and automate the process of obtaining medical
history in cases of Hymenoptera venom allergy [17].
Aesthetic Medicine
A text-based patient information chatbot named ‘Beautybot’ covering themes such as wrinkles and
pigmentation disorders. Written and image-based content was provided in a multiple-choice-driven
algorithmic workflow [18].

Chatbots have been examined with radiological [19] and histopathological languages,
with interesting translation results. Both studies reported the beneficial effect of chatbots
on those patients; the clinicians tried to understand the two different medical languages,
and the respective clinicians confirmed their usefulness. Regarding the reports on der-
matopathology, the majority of physicians either agreed or strongly agreed that ChatGPT-4
interpretations of the reports were thorough, precise, comprehensible, and unlikely to
result in harm, while there was no variance in physicians’ assessments of translations
between inflammatory and neoplastic diagnoses [13]. On the other hand, dermoscopy is a
rather complicated language, and to our knowledge, no study has examined the ability of a
chatbot to interpret dermoscopic descriptions.
Dermoscopy is a widely utilized noninvasive diagnostic method that enhances the
accuracy of diagnosing pigmented and cancer-suspicious lesions compared to naked eye
examination. The dermoscopy-related vocabulary is specialized. Many terms within this
lexicon are metaphorical, such as the “starburst” pattern, or signs such as the “delta
glider sign”. The descriptive terms are used to describe pigment distribution (for example,
dots and leaf-like structures), vessel morphology (arborizing and dotted vessels), and
keratin disturbances. Although descriptive terms and metaphors can assist in remembering
and establishing connections between particular dermoscopic features and diagnoses,
ambiguity stemming from identical descriptive terms for the same structure can pose a
potential obstacle to both learning and research within the field. As a result, it is reasonable
to wonder how a chatbot can deal with the understanding and creation of such specialized
vocabulary [20].
Diagnostics 2024, 14, 1165 4 of 21

Therefore, it is reasonable to assess the ability of a chatbot to enhance clinicians’ under-

standing and proficiency in dermoscopic examination through interactive, personalized,
and informative interactions.

2. Materials and Methods

This study employed a survey-based methodology to assess the performance and
utility of an artificial intelligence (AI) chatbot in translating dermoscopic descriptions of
skin lesions, supplemented with specific clinical information such as age, skin tone, and
immunocompetency. The AI chatbot utilized for this research was ChatGPT 3.5, recognized
as one of the most commonly used chatbots in everyday settings. The inclusion criteria
comprised individuals certified in the dermoscopic interpretation and diagnosis of skin
lesions (dermatologists or doctors with a certification in dermoscopy such as an MSc).
A structured survey was designed to evaluate the differential diagnosis provided
by the AI chatbot. The survey assessed the completeness and diagnostic utility of the
chatbot-generated reports, focusing on their contribution to accurate diagnosis and educa-
tional opportunities for novice dermatologists. Specific attention was given to potential
inaccuracies or misleading conclusions in the differential diagnosis. Participants were
presented with a series of hypothetical dermoscopic descriptions of skin lesions, each
accompanied by specific clinical information. They were subsequently requested to assess
the differential diagnosis list generated by the AI chatbot in its initial response, without
including the accompanying text or information, such as the descriptions of the listed
lesions or recommendations for visiting a dermatologist. Following each presentation, the
participants were prompted to provide feedback on the completeness, accuracy, and utility
of the chatbot-generated differential diagnosis. The correct answers were not revealed to
the participants at the time of evaluation.
Participation in this study was voluntary. The survey instrument ensured anonymity
of responses, and data confidentiality was strictly maintained throughout this study.
Each question was marked using a scale of 1 (disagree) to 3 (agree), and depending on
the answers of the clinician on each chatbot’s differential diagnosis, each case would obtain
an average score. The average scores on a questionnaire would be collected and compared
in each case. The survey comprised three instances of dermoscopic descriptions, covering
basal cell carcinoma (BCC), squamous cell carcinoma (SCC), benign lesions, pigmented
lesions, and inflammatory dermatoses in each case. It also included descriptions of signs,
along with three scenarios detailing age, lesion location, phenotype, and immune state
specifications. The total score comprised the average value and a standard deviation. The
average values were reported to one decimal place, while standard deviation values were
reported to two decimal places. The comparison of scores between more than two groups
was performed using a one-way analysis of variance test with post-hoc multiple two-group
comparisons for normal data, and a Kruskal–Wallis test was used for non-normal data. To
compare categorical data, the χ2 -test was performed. A Shapiro–Wilk test was used to test
for normality.

3. Results
A total of 30 clinicians with dermoscopic knowledge, who were eligible according
to our inclusion criteria, completed the questionnaire by judging the answers the chatbot
gave. The majority of participants (83.3%) were female professionals working across
various regions in Greece, including both rural and urban areas, as well as in hospitals and
health centers. Among them, 40% were clinicians with an MSc in dermoscopy, 20% were
dermatology residents, and 40% were experienced dermatologists with over twenty years
of clinical practice, collectively encompassing a broad spectrum of dermoscopic expertise.
The hypothetical dermoscopic scenarios included descriptions of skin lesions with
classic findings (Figure 1).
 ious regions in Greece, including both rural and urban areas, as well as in hospitals and
health centers. Among them, 40% were clinicians with an MSc in dermoscopy, 20% were
dermatology residents, and 40% were experienced dermatologists with over twenty years
of clinical practice, collectively encompassing a broad spectrum of dermoscopic expertise.
Diagnostics 2024, 14, 1165 The hypothetical dermoscopic scenarios included descriptions of skin lesions5with
of 21
classic findings (Figure 1).

Figure 1.
Figure 1. (A)
(A) Dermoscopy
DermoscopyofofaaBCC BCClesion
lesionincluding
including arborizing
arborizingvessels, ulceration,
vessels, ulceration,shiny white-red
shiny white-
structureless areas, and short white streaks (example used in Table 1). (B)
red structureless areas, and short white streaks (example used in Table 1). (B) Dermoscopy Dermoscopy image
imageof
seborrheic keratosis with fissures and ridges giving cerebriform appearance (example
of seborrheic keratosis with fissures and ridges giving cerebriform appearance (example used in used in ques-
tions regarding dermoscopic features of benign lesions). (C) Peripheral delicate pigment network
questions regarding dermoscopic features of benign lesions). (C) Peripheral delicate pigment network
and central white patch in case of dermatofibroma (used as a question to the chatbot). (D) Dermos-
and central white patch in case of dermatofibroma (used as a question to the chatbot). (D) Dermoscopy
copy image of psoriasis lesion with distributed dotted vessels in a reddish pink background and
image of psoriasis
white scale. lesion
(E) Yellow with
and distributed
black dots anddotted
at leastvessels in a reddish
one broken hair in pink
a casebackground
of alopecia and
areatawhite
(ex-
scale.
ample(E) Yellow
posed andchatbot
to the black dots andcategory
in the at least of
oneinflammatory
broken hair in a case of alopecia
dermatoses). areata (example
(F) Dermoscopy image
posed to theanchatbot
indicating in the
atypical category
nevus with anof inflammatory
atypical pigmentdermatoses).
network, (F) Dermoscopy
asymmetry image indicating
in structure or color,
brown
an dots,
atypical and irregular
nevus dots and
with an atypical globules.
pigment This image
network, was posed
asymmetry as a scenario
in structure to brown
or color, the chatbot, and
dots, and
the differential
irregular dots anddiagnosis was
globules. assessed.
This image was posed as a scenario to the chatbot, and the differential
diagnosis was assessed.
The initial set of queries to the chatbot comprised dermoscopic images associated
withTheBCC.initial settwo
Within of queries to the chatbot
of the descriptions comprised
provided, dermoscopic vascular
the characteristic images associated
pattern of
with BCC. Within two of the descriptions provided, the characteristic
BCC, known as arborizing vessels, and ulceration, a common feature in BCCs, were vascular pattern
ob-
of BCC,[21].
served knownBCCaswasarborizing
listed in vessels,
all of theand ulceration,
chatbot’s a common
differential feature
diagnoses. in BCCs, were
Additionally, the
observed [21].other
list included BCCskinwas cancers
listed inlike
all of
SCCtheand
chatbot’s differential
melanoma, as welldiagnoses.
as a moreAdditionally,
general cate-
the
gorylist included
termed other
“other skin cancersconditions”.
dermatological like SCC and melanoma,
However, as well also
the chatbot as aincluded
more general
diag-
category
noses thattermed “other dermatological
were inconsistent conditions”.
with the provided However,
description, suchtheaschatbot
atypicalalso included
nevus, as no
diagnoses
pigmentedthat were inconsistent
network was reported with the description.
in the provided description, suchvariations,
Despite these as atypicalthe
nevus,
lists as
in
no pigmented network was reported in the description. Despite these
all the examples received high scores from the assessing dermatologists (Table 1). variations, the lists
in all the examples received high scores from the assessing dermatologists (Table 1).
The next triad was about SCC-based dermoscopic images with the description involv-
ing classic dermoscopic features encountered in SCC such as the pink background, white
circles, and glomerular vessels. Also, different types of SCC were described as the second
description is in situ SCC (Bowen’s disease), and when the homogenous pigmentation is
included, the lesion is more likely to be a pigmented Bowen disease [22]. The chatbot’s
differential diagnosis list includes Bowen’s disease, along with pigmented BCC, actinic
keratosis, sebaceous hyperplasia, and other potential conditions. Amongst the descriptions,
no statistically significant differences were noticed performing Kruskal–Wallis statistics in
the “complete” (p = 0.128 > 0.05) category. In the category of “helpful to diagnosis,” a statis-
tically significant difference was observed between descriptions 1 and 3 (post-hoc Dunn’s
Diagnostics 2024, 14, 1165 6 of 21

test: p = 0.045 < 0.05), potentially because the provided description is broader and the
differential diagnosis included is more extensive. This pairing assists clinicians in making
diagnoses more efficiently. Moreover, the lack of correct answers in the list for the first SCC
description results in reduced scores for dermatologists. It is worth mentioning that both
the first and third descriptions stem from the good differentiation of SCC, indicating the
same type of SCC. A similar pattern of scores was followed in the “teaching tool” category
(p = 0.01 < 0.05), with statistical significance between the 1st and 2nd descriptions (post-hoc
Dunn’s test, p = 0.013) and 1st and 3rd descriptions (post-hoc Dunn’s test, p = 0.006) due to
the bad performance of the chatbot in the 1st description, not including the correct answer
(Table 2).

Table 1. BCC dermoscopy-based image scores according to the evaluation of dermatologists who
completed the questionnaire. The exact scores of the triad of the BCC dermoscopy-based images are
as follows: 1st description: Complete: 60% Agree, 10% Neither agree nor disagree, and 30% Disagree,
Total score: 2.3 ± 0.91, Helpful to diagnosis: 100% Agree, Total score: 3, Teaching tool: 100% agree,
and Total score: 3; 2nd description: Complete: 60% Agree, 30% Neither agree nor disagree, and
10% Disagree, Total score: 2.5 ± 0.68, Helpful to diagnosis: 90% Agree and 10% Neither agree nor
disagree, Total score: 2.9 ± 0.31, Teaching tool: 90% Agree and 10% Neither agree nor disagree, and
Total score: 2.9 ± 0.31; and 3rd description: Complete: 40% Agree, 30% Neither agree nor disagree,
and 30% Disagree, Total score: 2.1 ± 0.85, Helpful to diagnosis: 90% Agree and 10% Disagree, Total
score: 2.8 ± 0.61, Teaching tool: 90% Agree and 10% Disagree, and Total score: 2.8 ± 0.61.

Original Question to
the Chatbot:
What Is the Differential
Differential Diagnosis Generated Complete Helpful to Diagnosis Teaching Tool
Diagnosis of a Lesion
Whose Dermoscopic Image
Includes the Following?
Basal cell carcinoma (BCC)
1 (A) Arborizing vessels,
Squamous cell carcinoma (SCC)
(B) ulceration, (C) shiny
Amelanotic melanoma 2.3 ± 0.91 3 3
white-red structureless areas,
Pyogenic granuloma
and (D) short white streaks
Other dermatological conditions
Squamous cell carcinoma (SCC)
2 (A) Ulceration, (B) multiple Basal cell carcinoma (BCC)
small erosions, (C) shiny Pyogenic granuloma 2.5 ± 0.68 2.9 ± 0.31 2.9 ± 0.31
white-red structureless areas, Infectious or inflammatory
and (D) shiny white lines. dermatoses
Traumatic lesions
Melanoma
3 (A) Arborizing vessels,
Atypical nevus (dysplastic nevus)
(B) blue-grey ovoid nests
Basal cell carcinoma (BCC)
(large clods), (C) ulceration, 2.1 ± 0.85 2.8 ± 0.61 2.8 ± 0.61
Seborrheic keratosis
and (D) peppering (grey or
Dermatofibroma
brown dots) and /or clods.
Eccrine poroma
BCC dermoscopy-based
2.3 ± 0.83 2.9 ± 0.4 2.9 ± 0.4
images (Sum)

Benign lesion-based images triad were based on dermoscopic descriptions of ac-

tinic and seborrheic keratosis and dermatofibroma. The dermoscopic characteristics of
those are classic and frequently used in dermoscopic algorithms to exclude skin malig-
nancies [23–25]. The correct answer was included in all the answers in this triad, while
the differential diagnosis was wide, including inflammatory dermatoses, apart from skin
cancer and benign neoplasms (Table 3). Also, the 2nd description, which was represented
by seborrheic keratosis, did not include signs or vessels but a shape characterization (cere-
briform appearance). In this case as well, the chatbot found the correct answer and formed
Diagnostics 2024, 14, 1165 7 of 21

a satisfactory differential diagnosis according to the dermatologists-participants in the

survey. Statistical significance in the satisfaction of the participants was observed with
actinic keratosis achieving the lower score (Kruskal–Wallis test between the groups; groups:
complete: p = 0.03. help to diagnosis: p = 0, and teaching tool: p = 0). This indicates the
heterogeneity of the specific group as different benign lesions can have variably challenging
dermoscopic images.

Table 2. SCC dermoscopy-based image scores according to the evaluation of dermatologists who
completed the questionnaire. The exact scores of the triad of the SCC dermoscopy-based images are
as follows: 1st description: Complete: 50% Agree, 20% Neither agree nor disagree, and 30% Disagree,
Total score: 2.1 ± 0.85, Helpful to diagnosis: 60% Agree, 20% Neither agree nor disagree, and 20%
Disagree, Total score: 2.4 ± 0.81, Teaching tool: 50% Agree, 30% Neither agree nor disagree, and 20%
Disagree, and Total score: 2.3 ± 0.79; 2nd description: Complete: 50% Agree, 20% Neither agree nor
disagree, and 30% Disagree, Total score: 2.2 ± 0.89, Helpful to diagnosis: 80% Agree, 10% Neither
agree nor disagree, and 10% Disagree, Total score: 2.7 ± 0.65. Teaching tool: 80% Agree, 10% Neither
agree nor disagree, and 10% Disagree, and Total score: 2.7 ± 0.65; and 3rd description: Complete:
70% Agree, 10% Neither agree nor disagree, and 20% Disagree, Total score: 2.5 ± 0.82, Helpful to
diagnosis: 80% Agree and 20% Disagree, Total Score: 2.9 ± 0.3, Teaching tool: 80% Agree and 20%
Disagree, and Total score: 2.9 ± 0.3.

Original Question to
the Chatbot:
What Is the Differential
Differential Diagnosis Generated Complete Helpful to Diagnosis Teaching Tool
Diagnosis of a Lesion
Whose Dermoscopic Image
Includes the Following?
Lentigo
1 (A) Brown or grey dots,
Seborrheic keratosis
(B) pink or skin-colored
Melanocytic nevus
eccentric structureless areas, 2.1 ± 0.85 2.4 ± 0.81 2.3 ± 0.79
Dermatofibroma
and (C) glomerular
Amelanotic melanoma
(coiled) vessels
Angiokeratoma
Squamous cell carcinoma (SCC)
2 (A) A scaly surface, Bowen’s disease (squamous cell
(B) small brown globules, carcinoma in situ)
(C) linear greyish dots, and 2.2 ± 0.89 2.7 ± 0.65 2.7 ± 0.65
Lentigo maligna
(D) homogeneous Actinic keratosis
pigmentation Pigmented basal cell carcinoma (BCC)
Sebaceous hyperplasia
Seborrheic keratosis
3 (A) White circles, Squamous cell carcinoma (SCC)
(B) keratin, (C) pink Bowen’s disease
2.5 ± 0.82 2.9 ± 0.3 2.9 ± 0.3
background, and Verruca vulgaris (common wart)
(D) blood spots Pyogenic granuloma
Amelanotic melanoma
SCC dermoscopy-based
2.3 ± 0.86 2.6 ± 0.66 2.6 ± 0.67
images (Sum)

Pigmented lesions, especially lesions with a pigmented network on dermoscopy such

as nevi and melanoma, were assessed in the next triad. This triad aimed to test the chatbot
to determine if it can recognize a malignant lesion in the comparison between nevi (1st
and 3rd descriptions) and melanoma (2nd description). The 2nd description contained the
blue-black veil, which is a dermoscopic feature of melanoma [26,27] or pigmented BCC [28].
The differential diagnosis results included all melanocytic lesion subtypes (Table 4).
Diagnostics 2024, 14, 1165
Original Question to
the Chatbot:
What Is the Differential
Diagnosis of a Lesion
Whose Dermoscopic Image
Includes the Following?
1 (A) A pigmented
pseudo-network and
(B) widened follicular
openings along with
(C) fine scaling
2 Fissures and ridges giving
a cerebriform appearance
3 A peripheral delicate
pigment network and central
white patch
Benign lesion
dermoscopy-based
images (Sum)
8 of 21
Table 3. Benign lesion dermoscopy-based image scores according to the evaluation of dermatologists
who completed the questionnaire. The exact scores of the triad are as follows: 1st description
(actinic keratosis): Complete: 40% Agree, 30% Neither agree nor disagree, and 30% Disagree, Total
score: 2.1 ± 0.84, Helpful to diagnosis: 50% Agree and 50% Neither agree nor disagree, Total score:
2.5 ± 0.5, Teaching tool: 50% Agree, 40% Neither agree nor disagree, and 10% Disagree, and Total
score: 2.4 ± 0.67; 2nd description (seborrheic keratosis): Complete: 60% Agree, 20% Neither agree
nor disagree, and 20% Disagree, Total score: 2.4 ± 0.81, Helpful to diagnosis: 80% Agree and 20%
Neither agree nor disagree, Total score: 2.8 ± 0.4. Teaching tool: 80% Agree, 10% Neither agree
nor disagree, and 10% Disagree, and Total score: 2.7 ± 0.65; and 3rd description (dermatofibroma):
Complete: 70% Agree, 20% Neither agree nor disagree, and 10% Disagree, Total score: 2.6 ± 0.67,
Helpful to diagnosis: 100% Agree, Total score: 3, Teaching tool: 100% Agree, and Total score: 3.
Differential Diagnosis Generated Complete Helpful to Diagnosis Teaching Tool
Seborrheic keratosis
Lichen planus
Psoriasis
Lichen simplex chronicus 2.1 ± 0.84 2.5 ± 0.5 2.4 ± 0.67
Pityriasis rosea
Basal cell carcinoma (BCC)
Pigmented actinic keratosis
Seborrheic keratosis
Dermatosis papulosa nigra (DPN)
Verruca vulgaris (common wart)
2.4 ± 0.81 2.8 ± 0.4 2.7 ± 0.65
Squamous cell carcinoma (SCC)
Keratoacanthoma
Giant congenital melanocytic nevus
Melanocytic nevus
Dermatofibroma
Pigmented basal cell carcinoma (BCC)
Seborrheic keratosis
2.6 ± 0.67 3 3
Lentigo maligna
Post-inflammatory hypopigmentation
Halo nevus
Other melanocytic lesions
2.4 ± 0.8 2.8 ± 0.43 2.7 ± 0.59
Taking inflammatory dermatosis images into consideration, a case of psoriasis, sar-
coidosis, and alopecia areata was assessed. Psoriasis dermoscopy was presented with the
classic dermoscopic finding, with distributed dotted vessels in a reddish pink background
and white scales [29]. Additionally, the dermoscopy of sarcoidosis tends to be more chal-
lenging in clinical practice, with the orange structureless areas being characteristic of their
dermoscopic presentation [30]. Sarcoidosis was not included in the differential diagnosis of
the chatbot and was thus evaluated with low marks in completeness (2 ± 0.79) (Table 5).
Also, the third description was the first question to the chatbot that included scalp location.
The term “an alopecia plaque in the head” was added to that question, specifying the nature
of the disease. Statistical significance was found concerning the scores as a dermoscopic
image of psoriasis achieved a much higher score in two categories (Table 5) (Kruskal–Wallis
test between groups: complete: p = 0.02 and teaching tool: p = 0.007).
Diagnostics 2024, 14, 1165 9 of 21

Table 4. Pigmented lesion dermoscopy-based image scores according to the evaluation of derma-
tologists who completed the questionnaire. The exact scores of the triad of the pigmented lesion
dermoscopy-based images are as follows: 1st description: Complete: 70% Agree and 30% Disagree,
Total score: 2.2 ± 0.89, Helpful to diagnosis: 90% Agree and 10% Disagree, Total score: 2.7 ± 0.65,
Teaching tool: 80% Agree, 10% Neither agree nor disagree, and 10% Disagree, and Total score:
2.6 ± 0.67; 2nd description: Complete: 70% Agree, 10% Neither agree nor disagree, and 20% Dis-
agree, Total score: 2.5 ± 0.82, Helpful to diagnosis: 80% Agree and 20% Neither agree nor disagree,
Total score: 2.8 ± 0.4. Teaching tool: 80% Agree and 20% Neither agree nor disagree, and Total score:
2.8 ± 0.4; and 3rd description: Complete: 50% Agree, 20% Neither agree nor disagree, and 30%
Disagree, Total score: 2.2 ± 0.91, Helpful to diagnosis: 80% Agree, 10% Neither agree nor disagree,
and 10% Disagree, Total score: 2.7 ± 0.65, Teaching tool: 70% Agree, 20% Neither agree nor disagree,
and 10% Disagree, and Total score: 2.8 ± 0.4.

Original Question to the

Chatbot:
What Is the Differential
Differential Diagnosis Generated Complete Helpful to Diagnosis Teaching Tool
Diagnosis of a Lesion
Whose Dermoscopic Image
Includes the Following?
Melanoma
Atypical nevus
1 (A) An atypical pigment
Basal cell carcinoma (BCC)
network, asymmetry in
Squamous cell carcinoma (SCC)
structure or color, (B) brown 2.2 ± 0.89 2.7 ± 0.65 2.6 ± 0.67
Seborrheic keratosis
dots, and (C) irregular dots
Pigmented actinic keratosis
and globules
Atypical fibroxanthoma
Other malignancies
Melanoma
Basal cell carcinoma (BCC)
2 (A) A blue-white veil,
Pigmented actinic keratosis
(B) multiple brown dots, and 2.5 ± 0.82 2.8 ± 0.4 2.8 ± 0.4
Seborrheic keratosis
(C) pseudopods
Dermatofibroma
Spitzoid lesions
Melanocytic nevus
Lentigo
3 (A) Structureless Dysplastic nevus
hyperpigmented areas in the Melanoma 2.2 ± 0.91 2.7 ± 0.65 2.8 ± 0.4
center and (B) reticular at Pigmented basal cell carcinoma (BCC)
the periphery Pigmented Bowen’s disease
(squamous cell carcinoma in situ)
Amelanotic melanoma
Pigment network
dermoscopy 2.4 ± 0.88 2.8 ± 0.48 2.7 ± 0.59
lesion-based images

As indicated by the above evaluations, challenging dermoscopy can occur in the same
category of skin lesions. The scores produced cannot be only attributed to the excellence
in the differential diagnosis of the chatbot but also to the dermoscopic knowledge of the
participant. A clinician who has difficulties in finding the correct diagnosis is more likely
to select “neither agree nor disagree”. However, as illustrated in Table 6, which provides a
summary of the scores in each category, there was no statistical significance observed in
the “complete” category. For the Kruskal–Wallis score in this category between the groups,
p = 0.881 > 0.05. This suggests that the evaluation of chatbot differential diagnoses remains
consistently high, irrespective of the type of skin lesion. As for the diagnosis approach
and teaching tool usefulness, BCC scores outperformed SCC scores and inflammatory
dermatosis scores (post-hoc Dunn’s tests: p = 0.005 and 0 for the “diagnosis” category and
p = 0.001 and 0 for teaching tool usefulness, respectively). This outcome demonstrates
Diagnostics 2024, 14, 1165 10 of 21

the extensive familiarity dermatologists have with dermoscopic images of BCC while
also highlighting the complexity associated with interpreting dermoscopic images of SCC
and inflammatory dermatoses. In teaching mode, BCC also outperformed benign and
pigmented lesions (p = 0.004 both). Inflammatory dermatoses and SCC had the lowest
scores in the “teaching tool” category (Table 6). The total percentages of participants who
agreed, neither agreed nor disagreed, and disagreed in each skin category are also presented
as pie charts in Figures 2–4.

Table 5. Inflammatory dermatosis dermoscopy-based image scores according to the evaluation of

dermatologists who completed the questionnaire. The exact scores of the inflammatory dermatosis
dermoscopy-based images are as follows: 1st description (psoriasis): Complete: 80% Agree, 10%
Neither agree nor disagree, and 10% Disagree, Total score: 2.7 ± 0.65, Helpful to diagnosis: 80%
Agree, 10% Neither agree nor disagree, and 10% Disagree, Total score: 2.7 ± 0.65, Teaching tool: 80%
Agree, 10% Neither agree nor disagree, and 10% Disagree, and Total score: 2.7± 0.65; 2nd description
(sarcoidosis): Complete: 30% Agree, 40% Neither agree nor disagree, and 30% Disagree, Total score:
2 ± 0.79, Helpful to diagnosis: 50% Agree, 40% Neither agree nor disagree, and 10% Disagree, Total
score: 2.4 ± 0.67, Teaching tool: 40% Agree, 30% Neither agree nor disagree, and 30% Disagree, Total
score: 2.1 ± 0.85; and 3rd description (alopecia areata): Complete: 50% Agree, 20% Neither agree nor
disagree, and 30% Disagree, Total score: 2.2 ± 0.89, Helpful to diagnosis: 70% Agree, 20% Neither
agree nor disagree, and 10% Disagree, Total score: 2.6 ± 0.67, Teaching tool: 70% Agree, 10% Neither
agree nor disagree, and 20% Disagree, and Total score: 2.5 ± 0.82.

Original Question to
the Chatbot:
What Is the Differential
Differential Diagnosis Generated Complete Helpful to Diagnosis Teaching Tool
Diagnosis of a Lesion
Whose Dermoscopic Image
Includes the Following?
Psoriasis
Lichen planus
1 Distributed dotted vessels Discoid lupus erythematosus (DLE)
in a reddish pink Pityriasis rosea 2.7 ± 0.65 2.7 ± 0.65 2.7 ± 0.65
background and white scales Tinea corporis (ringworm)
Basal cell carcinoma (BCC)
Bowen’s disease
Seborrheic keratosis (SK)
Lichen planus
Basal cell carcinoma (BCC)
Angioma
2 Orange structureless areas
Hemangioma
with overlying focused linear 2 ± 0.79 2.4 ± 0.67 2.1 ± 0.85
Actinic keratosis (AK)
vessels and white areas
Bowen’s disease (squamous cell
carcinoma in situ)
Keratoacanthoma
Amelanotic melanoma
Dermatofibroma
Alopecia areata
Tinea capitis
Trichotillomania
3 Yellow and black dots and Discoid lupus erythematosus (DLE) 2.2 ± 0.89 2.6 ± 0.67 2.5 ± 0.82
at least one broken hair Central centrifugal cicatricial
alopecia (CCCA)
Secondary cicatricial alopecia
Folliculitis decalvans
Inflammatory dermatosis
2.3 ± 0.82 2.6 ± 0.67 2.4 ± 0.81
dermoscopy-based images
Diagnostics 2024, 14, 1165 11 of 21

Table 6. The table summarizes the scores for each dermoscopic image triad.

Complete Helpful to Diagnosis Teaching Tool Usefulness

BCC dermoscopy-based images (Sum) 2.3 ± 0.83 2.9 ± 0.4 2.9 ± 0.4
SCC dermoscopy-based images (Sum) 2.3 ± 0.86 2.6 ± 0.66 2.6 ± 0.67
Benign lesion dermoscopy-based images (Sum) 2.4 ± 0.8 2.8 ± 0.43 2.7 ± 0.59
Pigment network2024,
Diagnostics dermoscopy
14, x FORlesion-based
PEER REVIEW 12 of 22
2.4 ± 0.88 2.8 ± 0.48 2.7 ± 0.59
images (Sum)
Inflammatory dermatosis dermoscopy-based
2.3 ± 0.82 2.6 ± 0.67 2.4 ± 0.81
images (Sum)

Pie charts
Figure 2.Figure 2. Pieshowing the percentage
charts showing of participants
the percentage who agreed,
of participants neither neither
who agreed, agreed agreed
nor dis-nor disa-
greed,
agreed, and and disagreed
disagreed regarding
regarding the completeness
the completeness of the chatbot’s
of the chatbot’s differential
differential diagnosisdiagnosis
based onbased
the on the
dermoscopic description for each category of
dermoscopic description for each category of skin lesions.skin lesions.

Table 6 illustrates the summary of scores in each dermoscopic image triad.

The total percentages of participants who agreed, neither agreed nor disagreed, and
disagreed in each skin category are also presented as pie charts in Figures 2–4.
Apart from classic descriptions of dermoscopic images, there was a triad of questions
that presented signs in dermoscopy with descriptive language and not with their adap-
tation terms. For example, the so-called rosettes found in actinic keratosis and SCC were
presented as four white points arranged as a four-leaf clover. Also, delta signs, commonly
found in scabies, were presented as multiple brownish triangular structures. Finally, a
strawberry pattern, commonly seen in actinic hyperkeratosis, was described as background
erythema interrupted by multiple small keratin-filled follicular ostia [31]. All three cate-
gories collected low scores on all the questions (Table 7). This highlights the significance of
the information provided by clinicians to the AI program being utilized, such as a chatbot.
Moreover, the lower scores may be attributed to the limited dermoscopic information upon
which the chatbot seemed to base its diagnostic list. For instance, it appeared that the
chatbot determined its diagnoses based on each term provided, such as the colors men-
tioned. For example, when the term “brownish” was included in the description of the delta
sign, it seemed to prompt the chatbot to generate diagnoses associated with dermoscopic
features containing this color, often indicative of melanocytic lesions. A similar approach
was observed with other responses as well, with terms like “keratin” leading to diagnoses
related to seborrheic keratosis. Comparing those findings with the previous summary

Figure 3. Pie charts illustrating the proportions of participants who agreed, neither agreed nor dis-
agreed, and disagreed regarding the “helpful to diagnosis” category of the chatbot’s differential
diagnosis based on the dermoscopic description for each skin lesion.
Diagnostics 2024, 14, 1165 12 of 21
Figure 2. Pie charts showing the percentage of participants who agreed, neither agreed nor disa-
greed, and disagreed regarding the completeness of the chatbot’s differential diagnosis based on the
dermoscopic description for each category of skin lesions.
scores, the descriptive terms did not collect enough points (Kruskal–Wallis testing between
all groups with p < 0.01).

Diagnostics 2024, 14, x FOR PEER REVIEW 13 of 22

Figure3.3. Pie
Figure Pie charts illustrating the
charts illustrating the proportions
proportionsofofparticipants
participantswho
whoagreed,
agreed, neither
neither agreed
agreed nornor
dis-
disagreed, and disagreed regarding
agreed, and disagreed regarding the the “helpful to diagnosis” category of the chatbot’s differential
diagnosis” category of the chatbot’s differential
diagnosis
diagnosisbased
basedononthe
thedermoscopic
dermoscopic description forfor
description each skin
each lesion.
skin lesion.

Figure 4. Pie
Figure 4.charts showing
Pie charts the proportions
showing of participants
the proportions who agreed,
of participants neither agreed
who agreed, neithernor disagreed,
agreed nor disa-
andgreed,
disagreed
andregarding
disagreedhow useful how
regarding the chatbot’s differential
useful the chatbot’sdiagnosis teaching
differential toolteaching
diagnosis can be based on be
tool can
based on the dermoscopic
the dermoscopic description
description for for each skin lesion.
each skin lesion.

Apart from classic descriptions of dermoscopic images, there was a triad of questions
that presented signs in dermoscopy with descriptive language and not with their adapta-
tion terms. For example, the so-called rosettes found in actinic keratosis and SCC were
presented as four white points arranged as a four-leaf clover. Also, delta signs, commonly
found in scabies, were presented as multiple brownish triangular structures. Finally, a
strawberry pattern, commonly seen in actinic hyperkeratosis, was described as back-
ground erythema interrupted by multiple small keratin-filled follicular ostia [31]. All three
categories collected low scores on all the questions (Table 7). This highlights the signifi-
Diagnostics 2024, 14, 1165 13 of 21

Table 7. Table presenting cases based on dermoscopic signs.

Original Question to
the Chatbot:
What Is the Differential
Differential Diagnosis Generated Complete Helpful to Diagnosis Teaching Tool
Diagnosis of a Lesion
Whose Dermoscopic Image
Includes the Following?
Dermatofibroma
Seborrheic keratosis
Four white points arranged Lichen planus-like keratosis (LPLK)
1.7 ± 0.79 2 ± 0.79 2 ± 0.79
as a four-leaf clover Angiokeratoma
Basal cell carcinoma (BCC)
Melanocytic lesions
Melanocytic nevus
Multiple brownish triangular Reed nevus
structures creating a Spitzoid lesions 1.6 ± 0.76 1.9 ± 0.84 2 ± 0.79
delta sign Lentigo
Melanoma
Seborrheic keratosis
Background erythema Dermatosis papulosa nigra (DPN)
interrupted by multiple Angioma serpiginosum
1.7 ± 0.92 2.2 ± 0.76 2.1 ± 0.84
small keratin-filled Lichen planus-like keratosis (LPLK)
follicular ostia Follicular keratosis
Basal cell carcinoma (BCC)
Signs and descriptions (Sum) 1.7 ± 0.82 2 ± 0.8 2 ± 0.8

Table 7 shows three dermoscopic signs with an image-based evaluation. The 1st and
3rd descriptions represent rosettes and strawberry patterns observed in actinic keratosis,
while the 2nd dermoscopic image is based on the delta sign seen in scabies.
The subsequent triads integrate dermoscopic images, along with an additional patient
characteristic, such as age, phototype, or immune state. This encompasses scenarios com-
paring patients aged 20 versus 70 years, patients with lighter and darker phototypes, and
immunocompetent patients and individuals under immunosuppression (Tables 8–10). The
differential diagnosis list is adapted to include lesion types appropriate for each category.
For instance, in the case of elderly patients, the differential diagnosis included more skin
cancer subtypes and lesions like seborrheic and actinic keratosis, which are commonly
observed in this demographic [32]. Additionally, atypical spitz nevus, often observed in
younger patients and alert clinicians, was also noted [33]. In the third description, the
chatbot-generated list remained the same for both the young and old patients (Table 8),
indicating that the diagnostic approach adopted by the chatbot remains consistent in some
cases. In the case of dark-skinned patients, pigmented lesions such as pigmented BCC,
pigmented actinic keratosis, and keratosis papulosa nigra keratosis, as well as skin cancer
subtypes commonly reported in black individuals, such as dermatofibrosarcoma and acral
melanoma, were more common [34]. In the case of immunosuppression, cases of Kaposi
sarcoma, cutaneous lymphoma, and atypical infections were commonly reported, following
the incidence of skin diseases in immunosuppressed patients [35].
Table 8 shows the evaluation scores of the chatbot differential diagnosis based on the
dermoscopic images plus the characteristics of age.
Table 9 shows the evaluation scores of the chatbot differential diagnosis based on the
dermoscopic images plus the characteristics of phenotype.
Table 10 shows the evaluation scores of the chatbot differential diagnosis based on the
dermoscopic images plus the characteristics of the immune state.
Diagnostics 2024, 14, 1165 14 of 21

Table 8. Table displaying the evaluation scores for the chatbot’s differential diagnosis, considering dermoscopic images and age characteristics.

Original Question to
the Chatbot:
Dermoscopy What Is the Differential
Category Diagnosis of a Lesion Whose
Dermoscopic Image Includes
the Following?
(A) Arborizing vessels,
(B) ulceration, (C) shiny
BCC
white-red structureless areas,
and (D) short white streaks
(A) Brown or grey dots,
(B) pink or skin-colored
SCC eccentric structureless areas,
and (C) glomerular
(coiled) vessels
An atypical pigment network,
asymmetry in structure or
Pigmented lesion
color, brown dots, and
irregular dots and globules
Differential Diagnosis Differential Diagnosis Helpful to
Complete Teaching Tool
Generated (20 Years) Generated (70 Years Old) Diagnosis
Melanoma Melanoma
Atypical spitz nevus Basal cell carcinoma (BCC)
Basal cell carcinoma (BCC) Squamous cell carcinoma a (SCC)
Seborrheic keratosis Amelanotic melanoma 2.3 ± 0.79 2.8 ± 0.61 2.7 ± 0.65
Amelanotic melanoma Pyogenic granuloma
Pyogenic granuloma Seborrheic keratosis
Nodular melanoma Actinic keratosis
Melanocytic nevus Seborrheic keratosis
Seborrheic keratosis Lentigo
Lentigo Melanocytic nevus
2.2 ± 0.76 2.6 ± 0.67 2.6 ± 0.67
Dermatofibroma Dermatofibroma
Angiokeratoma Angiokeratoma
Amelanotic melanoma Amelanotic melanoma
Melanoma Melanoma
Atypical nevus Atypical nevus
Basal cell carcinoma (BCC) Basal cell carcinoma (BCC)
Squamous cell carcinoma (SCC) Squamous cell carcinoma (SCC) 2.3 ± 0.92 2.7 ± 0.65 2.7 ± 0.65
Seborrheic keratosis Seborrheic keratosis
Pigmented actinic keratosis Pigmented actinic keratosis
Lentigo maligna Lentigo maligna
Diagnostics 2024, 14, 1165 15 of 21

Table 9. Table displaying the evaluation scores for the chatbot’s differential diagnosis, considering dermoscopic images and phenotypic characteristics.

Original Question to
the Chatbot:
Dermoscopy What Is the Differential
Category Diagnosis of a Lesion Whose
Dermoscopic Image Includes
the Following?
(A) Arborizing vessels,
(B) ulceration, (C) shiny
BCC
white-red structureless areas,
and (D) short white streaks
(A) Brown or grey dots,
(B) pink or skin-colored
SCC eccentric structureless areas,
and (C) glomerular
(coiled) vessels
An atypical pigment network,
asymmetry in structure or
Pigmented lesion
color, brown dots, and
irregular dots and globules
Differential Diagnosis
Differential Diagnosis Generated Helpful to
Generated (without Any Complete Teaching Tool
(with Darker Phenotypes) Diagnosis
Characteristics)
Melanoma
Basal cell carcinoma (BCC) Basal cell carcinoma (BCC)
Squamous cell carcinoma (SCC) Squamous cell carcinoma (SCC)
amelanotic melanoma Hemangioma 2.6 ± 0.67 2.9 ± 0.31 2.9 ± 0.31
Pyogenic granuloma Pigmented basal cell carcinoma (PBCC)
Other dermatological conditions Dermatofibroma
Traumatic or inflammatory lesions
Seborrheic keratosis Dermatosis
Lentigo
papulosa nigra
Seborrheic keratosis
Angiokeratoma
Melanocytic nevus
LichenpPlanus 2.5 ± 0.82 2.9 ± 0.31 2.7 ± 0.65
dermatofibroma amelanotic
Acral lentiginous
melanoma
Melanoma
Angiokeratoma
Pigmented basal cell carcinoma (PBCC)
Melanoma
Melanoma
Atypical nevus
Atypical nevus
Basal cell carcinoma (BCC)
Basal cell carcinoma (BCC)
Squamous cell carcinoma (SCC)
Squamous cell carcinoma (SCC) 2.5 ± 0.82 2.7 ± 0.65 2.6 ± 0.67
Seborrheic keratosis
Seborrheic keratosis
Pigmented actinic keratosis
Pigmented actinic keratosis
Atypical fibroxanthoma
Lentigo maligna
Other malignancies
Diagnostics 2024, 14, 1165
Table 10. Table displaying the evaluation scores for the chatbot’s differential diagnosis, taking into account dermoscopic images and immune status characteristics.
Original Question to
the Chatbot:
Dermoscopy What Is the Differential
Category Diagnosis of a Lesion Whose
Dermoscopic Image Includes
the Following?
(A) Arborizing vessels,
(B) ulceration, (C) shiny
BCC
white-red structureless areas,
and (D) short white streaks
(A) Brown or grey dots,
(B) pink or skin-colored
SCC eccentric structureless areas,
and (C) glomerular (coiled)
vessels
An atypical pigment network,
asymmetry in structure or
Pigmented lesion
color, brown dots, and
irregular dots and globules
16 of 21
Differential Diagnosis
Differential Diagnosis Generated Helpful to
Generated (without Any Complete Teaching Tool
(with Immunosuppression) Diagnosis
Characteristics)
Cutaneous lymphoma
Kapos sarcoma
Basal cell carcinoma (BCC)
Atypical infections
Squamous cell carcinoma (SCC)
Aggressive squamous cell
Amelanotic melanoma 2.7 ± 0.65 3 2.7 ± 0.65
carcinoma (SCC)
Pyogenic granuloma
Atypical melanoma
Other dermatological conditions
Merkel cell carcinoma
Reactivation of viral infections
Kaposi sarcoma
Lentigo
Cutaneous lymphoma
Seborrheic keratosis
Atypical infections
Melanocytic nevus
Aggressive squamous cell 2.2 ± 0.76 2.7 ± 0.47 2.5 ± 0.51
Dermatofibroma amelanotic
carcinoma (SCC)
melanoma
Atypical melanoma
Angiokeratoma
Merkel cell carcinoma
Melanoma
Melanoma
Atypical nevus
Atypical nevus
Basal cell carcinoma (BCC)
Basal cell carcinoma (BCC)
Squamous cell carcinoma (SCC)
Squamous cell carcinoma (SCC) 2.5 ± 0.68 3 2.8 ± 0.4
Seborrheic keratosis
Kaposi sarcoma
Pigmented actinic keratosis
Cutaneous lymphoma
Atypical fibroxanthoma
Atypical infections
Other malignancies
Diagnostics 2024, 14, 1165 17 of 21

Completeness of the chatbot differential diagnosis: When dark skin and immune state
information were combined with BCC dermoscopy, higher scores were achieved compared
to dermoscopic descriptions alone, as well as when combined with age information. How-
ever, non-statistically significant results were achieved (Kruskal–Wallis test: p = 0.191)
between groups with and without such information. For SCC, the dark skin category
yielded the highest scores among specific SCC dermoscopic descriptions without statistical
significance (p = 0.162). Additionally, when focusing on a pigmented lesion description
centered around an atypical nevus, no statistically significant differences were observed
(p = 0.841).
Helpful to diagnosis: BCC descriptions across all categories (with or without additional
parameters) received consistently high scores, with no statistically significant differences
between them (p = 0.1). Incorporating parameters in SCC dermoscopic descriptions resulted
in statistically significant higher scores compared to descriptions alone (p = 0.040), especially
between simple descriptions and descriptions with skin type data (p = 0.004). A similar
trend was not observed for pigmented lesions (p = 0.052).
Teaching tool usefulness: All categories and lesions received high scores, with few
exceptions such as SCC and pigmented lesions in dark-skinned patients and SCC in patients
with immunosuppression. No statistically significant difference was reported between the
groups (BCC: p = 0.052, SCC: p = 0.097, and pigmented lesion: p = 0.997).

4. Discussion
This survey marks the initial attempt to evaluate the comprehensibility of the dermo-
scopic language by a chatbot, shedding light on the interaction between dermatologists
and AI programs in this specific realm of dermatology. Even among dermatologists, the
language used in dermoscopy can be perplexing as descriptive and metaphorical lan-
guages are occasionally employed, leading to a subjective interpretation of the observed
image [31]. Furthermore, the diagnostic utility of a dermoscopic image may vary based
on the equipment accessible and the expertise of each participant. Hence, dermoscopy
presents a challenging subject for assessment when integrating AI technologies.
The first step of our study was to evaluate the differential diagnosis list when examples
of different skin lesions were presented. The examples included dermoscopic descriptions
of skin cancers such as BCC, SCC, and melanoma, skin cancer mimickers such as actinic and
seborrheic keratosis, dermatofibroma, and atypical nevus, and inflammatory dermatoses
such as psoriasis and alopecia areata. Each participant evaluated each description on a
scale of 1 to 3 based on how complete, helpful to diagnosis, and educational the given
differential list was.
It is worth mentioning that when the chatbot was asked about a differential diagnosis,
except for the list presented in the previous tables, it reported an accompanying text. This
text usually presented extra features to look for when assessing a lesion that could lead
to the diagnosis. For example, in the case of a psoriasis description, the original question
to the chatbot was “what is the differential diagnosis of a lesion whose dermoscopy in-
cludes distributed dotted vessels in a reddish pink background and white scales”. The
chatbot began to generate its diagnosis list, assuming a probability priority. Additionally, it
associated other diagnoses with additional characteristics that, when combined with the
provided description, may alter the diagnosis while prompting the user to seek out those
characteristics. For instance, it links a violaceous background with lichen planus [36], a
central white scar-like area with dermatofibroma [37], and ulceration with Bowen’s dis-
ease [38]. As a result, it creates an algorithm based on the description given and potentially
newly discovered characteristics.
Questions regarding slight changes in the questions to the chatbot could lead to
a different differential diagnosis. The dermoscopic language has its own vocabulary;
therefore, synonyms can be used. For example, in the case of BCC, instead of “arborizing”,
the term “branching” can be used to describe the vessels of the skin cancer. In this case, a
similar differential diagnosis is produced (BCC, SCC, melanoma, etc.). The same occurs if a
Diagnostics 2024, 14, 1165 18 of 21

non-dermoscopic term is replaced and another adjective is used. For example, if the term
“scattered” is used to describe the dotted vessels in the psoriasis dermoscopy, the chatbot
maintains a similar set of potential diagnoses (psoriasis, lichen planus, and pityriasis rosea).
Furthermore, all the responses to the user concluded with a message advocating for a
visit to a clinician or a specialized dermatologist, along with highlighting the likelihood of
undergoing a biopsy for the ultimate diagnosis.
Completeness: In each case, the chatbot produced a long list of differential diagnoses,
scoring low in the case of SCC and inflammatory dermatoses but not with statistical
significance, meaning that the participants were almost equally satisfied with the answers
given. The scores were reduced in the case of practical descriptions of dermoscopic signs,
such as the description of the delta sign as multiple brownish triangular structures and the
strawberry pattern as background erythema interrupted by multiple small keratin-filled
follicular ostia. In those cases, the chatbot did not produce a satisfactory answer, maybe
due to the subjective use of the dermoscopic language and the lack of key dermoscopy
words that could easily reveal the answer. Incorporating patient information such as age,
phototype, and immune state did not alter the scores in the BCC and pigmented lesion
categories, but it did slightly increase scores for SCC without reaching statistical significance.
This suggests that in complex cases, clinicians may require additional available information.
Diagnosis approach: The diagnostic approach category garnered more points com-
pared to the complete category. Additionally, we posit that presenting the entire response
to participants, including the added characteristics to consider for assessing alternative di-
agnoses, would likely result in higher scores. In this instance, the scores for BCC surpassed
those for SCC and inflammatory dermatoses (post-hoc Dunn’s tests: p = 0.005 and 0 for
the diagnosis category). This outcome underscores dermatologists’ extensive familiarity
with dermoscopic images of BCC while also underscoring the challenges associated with
interpreting dermoscopic images of SCC and inflammatory dermatoses. Incorporating pa-
tient information such as age, phototype, and immune state did not alter the scores in most
cases. However, the diagnostic list generated from the description of SCC in a dark-skinned
patient received more points than the description of SCC alone. This may be because SCC
in dark-skinned patients is rare and may exhibit unique dermoscopic characteristics [39].
Additionally, the list generated in this case adhered to the epidemiology of cancerous
lesions in dark-skinned patients [40], potentially further satisfying the participants.
Teaching method: In this category, the chatbot answers collected the most points, indi-
cating that AI can not only help in clinical practice but can also be a valuable educational
tool. As in the diagnosis category, the scores for BCC surpassed those for SCC and inflam-
matory dermatoses (post-hoc Dunn’s tests: p = 0.001 and 0 for teaching tool usefulness,
respectively). It is worth noting that the scores for the “teaching” method usually aligned
with the scores for the “helpful to diagnosis” method, indicating that participants believe
that a tool that aids in diagnosis can be highly educational for medical students and new
clinicians. In cases where patients’ demographic data were added, SCC and pigmented
lesions in dark-skinned patients and SCC in patients with immunosuppression received
the lowest scores, but no statistical significance between groups was reported. This proves
that adding characteristics of the patients did not alter the quality of the chatbot’s answers.
A study with a similar method evaluated the ability of chatbots to recognize the der-
matopathological language, similar to how we assessed the dermoscopic language [13]. Der-
matopathological language as a dermoscopic language uses its own specialized vocabulary
and terminology, including cell types, cancer infiltration levels, and immunohistochemical
stains. In our study, the chatbots were assessed by experts regarding completeness, accu-
racy, ease of understanding, the possibility of causing stress to patients, and the likelihood
of causing harm. This study is more focused on the patient’s use of chatbots. It is worth
mentioning that BCC received one of the highest scores for the “ease of understanding”
category, and melanoma scored lowest in the “likelihood to cause harm” category. It is
important to recognize that dermoscopy and dermatopathology use different terminologies,
resulting in varying levels of chatbot comprehension regarding the same diagnosis.
Diagnostics 2024, 14, 1165 19 of 21

Chatbots and, generally, AI apps, as indicated by the abovementioned results, can

offer support, particularly in scenarios where a clinician might overlook a diagnosis in a
dermoscopic image requiring assessment. This collaboration becomes crucial in situations
where dermoscopic expertise is limited, time is constrained, or when handling a large
volume of cases. Thus, apart from focusing on comparing clinicians and AI applications,
emphasis should be placed on the outcomes of their collaboration. Chatbots serve as
beneficial tools in healthcare settings, and they cannot substitute for the expertise and
judgment of trained medical specialists [8].
Limitations: Our study has several limitations that warrant acknowledgment. First,
it relies on the ChatGPT3.5 chatbot, chosen due to its accessibility to the general popula-
tion. However, employing more advanced chatbots could yield different results. Second,
the differential diagnosis list in each case is derived from the initial response generated
by the chatbot. Also, the answers depend on the input information related to dermo-
scopic knowledge. Considering that subsequent responses may vary slightly, this could
potentially impact the outcomes. Finally, the survey was conducted among dermatologists
for evaluation, indicating potential variations in participants’ levels of experience and
dermoscopic knowledge.

5. Conclusions
The AI chatbot underwent testing to provide diagnoses based on the dermoscopic
language, and it performed well, generating mostly comprehensive, diagnostically helpful,
and educational lists of differential diagnoses. The evaluation of the chatbot was conducted
using dermoscopic descriptions of skin cancers like BCC, SCC, and melanoma, skin cancer
mimickers such as actinic and seborrheic keratosis, dermatofibroma, and atypical nevus,
and inflammatory dermatoses like psoriasis and alopecia areata. Lower scores were ob-
served for the AI chatbot in more challenging cases of SCC and inflammatory dermatoses
and in the assessment of subjective descriptions of dermoscopic signs. Furthermore, by in-
corporating patient characteristics such as age, phototype, or immune state, the differential
diagnosis list in each case was tailored to include lesion types suitable for each category,
highlighting the AI’s adaptability in assessing diagnoses and making it a valuable aid
for dermatologists.

Author Contributions: E.K. (Emmanouil Karampinis), O.T., K.-E.G., E.K. (Elli Kampra), C.S. and E.Z.
contributed to the writing of the manuscript and information collection. A.-V.R.S. and E.Z. reviewed
the final submission of the manuscript. All authors have read and agreed to the published version of
the manuscript.
Funding: This study received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data described in this study are available upon request from the
corresponding author.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Omiye, J.A.; Gui, H.; Daneshjou, R.; Cai, Z.R.; Muralidharan, V. Principles, Applications, and Future of Artificial Intelligence in
Dermatology. Front. Med. 2023, 10, 1278232. [CrossRef] [PubMed]
2. Li, Z.; Koban, K.C.; Schenck, T.L.; Giunta, R.E.; Li, Q.; Sun, Y. Artificial Intelligence in Dermatology Image Analysis: Current
Developments and Future Trends. J. Clin. Med. 2022, 11, 6826. [CrossRef]
3. Foltz, E.A.; Witkowski, A.; Becker, A.L.; Latour, E.; Lim, J.Y.; Hamilton, A.; Ludzik, J. Artificial Intelligence Applied to Non-
Invasive Imaging Modalities in Identification of Nonmelanoma Skin Cancer: A Systematic Review. Cancers 2024, 16, 629.
[CrossRef]
4. Martin-Gonzalez, M.; Azcarraga, C.; Martin-Gil, A.; Carpena-Torres, C.; Jaen, P. Efficacy of a Deep Learning Convolutional Neural
Network System for Melanoma Diagnosis in a Hospital Population. Int. J. Environ. Res. Public Health 2022, 19, 3892. [CrossRef]
Diagnostics 2024, 14, 1165 20 of 21

5. Karampinis, E.; Papadopoulou, M.-M.; Chaidaki, K.; Georgopoulou, K.-E.; Magaliou, S.; Roussaki Schulze, A.V.; Bogdanos, D.P.;
Zafiriou, E. Plaque Psoriasis Exacerbation and COVID-19 Vaccination: Assessing the Characteristics of the Flare and the Exposome
Parameters. Vaccines 2024, 12, 178. [CrossRef] [PubMed]
6. Du, A.X.; Ali, Z.; Ajgeiy, K.K.; Dalager, M.G.; Dam, T.N.; Egeberg, A.; Nissen, C.V.S.; Skov, L.; Thomsen, S.F.; Emam, S.; et al.
Machine Learning Model for Predicting Outcomes of Biologic Therapy in Psoriasis. J. Am. Acad. Dermatol. 2023, 88, 1364–1367.
[CrossRef] [PubMed]
7. Altamimi, I.; Altamimi, A.; Alhumimidi, A.S.; Altamimi, A.; Temsah, M.-H. Artificial Intelligence (AI) Chatbots in Medicine: A
Supplement, Not a Substitute. Cureus 2023, 15, e40922. [CrossRef] [PubMed]
8. Chakraborty, C.; Pal, S.; Bhattacharya, M.; Dash, S.; Lee, S.-S. Overview of Chatbots with Special Emphasis on Artificial
Intelligence-Enabled ChatGPT in Medical Science. Front. Artif. Intell. 2023, 6, 1237704. [CrossRef]
9. Kim, Y.-H.; Zhang, M.Z.; Vidal, N.Y. ChatGPT Offers an Editorial on the Opportunities for Chatbots in Dermatologic Research
and Patient Care. Dermatol. Online J. 2024, 29, 2. [CrossRef] [PubMed]
10. Diamond, C.; Rundle, C.W.; Albrecht, J.M.; Nicholas, M.W. Chatbot Utilization in Dermatology: A Potential Amelioration to
Burnout in Dermatology. Dermatol. Online J. 2022, 28, 16. [CrossRef] [PubMed]
11. Walss, M.; Anzengruber, F.; Arafa, A.; Djamei, V.; Navarini, A.A. Implementing Medical Chatbots: An Application on Hidradenitis
Suppurativa. Dermatology 2021, 237, 712–718. [CrossRef]
12. O’Hagan, R.; Poplausky, D.; Young, J.N.; Gulati, N.; Levoska, M.; Ungar, B.; Ungar, J. The Accuracy and Appropriateness of
ChatGPT Responses on Nonmelanoma Skin Cancer Information Using Zero-Shot Chain of Thought Prompting. JMIR Dermatol.
2023, 6, e49889. [CrossRef]
13. Zhang, Y.; Chen, R.; Nguyen, D.; Choi, S.; Gabel, C.; Leonard, N.; Yim, K.; O’Donnell, P.; Elaba, Z.; Deng, A.; et al. Assessing
the Ability of an Artificial Intelligence Chatbot to Translate Dermatopathology Reports into Patient-Friendly Language: A
Cross-Sectional Study. J. Am. Acad. Dermatol. 2024, 90, 397–399. [CrossRef] [PubMed]
14. Lam Hoai, X.-L.; Simonart, T. Comparing Meta-Analyses with ChatGPT in the Evaluation of the Effectiveness and Tolerance of
Systemic Therapies in Moderate-to-Severe Plaque Psoriasis. J. Clin. Med. 2023, 12, 5410. [CrossRef]
15. Lakdawala, N.; Channa, L.; Gronbeck, C.; Lakdawala, N.; Weston, G.; Sloan, B.; Feng, H. Assessing the Accuracy and Comprehen-
siveness of ChatGPT in Offering Clinical Guidance for Atopic Dermatitis and Acne Vulgaris. JMIR Dermatol. 2023, 6, e50409.
[CrossRef] [PubMed]
16. Yan, S.; Du, D.; Liu, X.; Dai, Y.; Kim, M.-K.; Zhou, X.; Wang, L.; Zhang, L.; Jiang, X. Assessment of the Reliability and Clinical
Applicability of ChatGPT’s Responses to Patients’ Common Queries About Rosacea. Patient Prefer. Adherence 2024, 18, 249–253.
[CrossRef] [PubMed]
17. Schneider, S.; Gasteiger, C.; Wecker, H.; Höbenreich, J.; Biedermann, T.; Brockow, K.; Zink, A. Successful Usage of a Chatbot to
Standardize and Automate History Taking in Hymenoptera Venom Allergy. Allergy 2023, 78, 2526–2528. [CrossRef]
18. Feuchter, S.; Kunz, M.; Djamei, V.; Navarini, A.A. Anonymous Automated Counselling for Aesthetic Dermatology Using a
Chatbot—An Analysis of Age- and Gender-specific Usage Patterns. J. Eur. Acad. Dermatol. Venereol. 2021, 35, E194–E195. [CrossRef]
19. Jeblick, K.; Schachtner, B.; Dexl, J.; Mittermeier, A.; Stüber, A.T.; Topalis, J.; Weber, T.; Wesp, P.; Sabel, B.O.; Ricke, J.; et al. ChatGPT
Makes Medicine Easy to Swallow: An Exploratory Case Study on Simplified Radiology Reports. Eur. Radiol. 2023, 34, 2817–2825.
[CrossRef]
20. Kittler, H.; Marghoob, A.A.; Argenziano, G.; Carrera, C.; Curiel-Lewandrowski, C.; Hofmann-Wellenhof, R.; Malvehy, J.;
Menzies, S.; Puig, S.; Rabinovitz, H.; et al. Standardization of Terminology in Dermoscopy/Dermatoscopy: Results of the Third
Consensus Conference of the International Society of Dermoscopy. J. Am. Acad. Dermatol. 2016, 74, 1093–1106. [CrossRef]
21. Lallas, A.; Apalla, Z.; Argenziano, G.; Longo, C.; Moscarella, E.; Specchio, F.; Raucci, M.; Zalaudek, I. The Dermatoscopic Universe
of Basal Cell Carcinoma. Dermatol. Pract. Concept. 2014, 4, 11–24. [CrossRef]
22. Papageorgiou, P.P.; Koumarianou, A.; Chu, A.C. Pigmented Bowen’s Disease. Br. J. Dermatol. 1998, 138, 515–518. [CrossRef]
23. Reinehr, C.P.H.; Bakos, R.M. Actinic Keratoses: Review of Clinical, Dermoscopic, and Therapeutic Aspects. An. Bras. Dermatol.
2019, 94, 637–657. [CrossRef] [PubMed]
24. Lin, J.; Han, S.; Cui, L.; Song, Z.; Gao, M.; Yang, G.; Fu, Y.; Liu, X. Evaluation of Dermoscopic Algorithm for Seborrhoeic Keratosis:
A Prospective Study in 412 Patients. J. Eur. Acad. Dermatol. Venereol. 2014, 28, 957–962. [CrossRef] [PubMed]
25. Zaballos, P.; Puig, S.; Llambrich, A.; Malvehy, J. Dermoscopy of Dermatofibromas. Arch. Dermatol. 2008, 144, 75–83. [CrossRef]
[PubMed]
26. Carrera, C.; Segura, S.; Aguilera, P.; Scalvenzi, M.; Longo, C.; Barreiro, A.; Broganelli, P.; Cavicchini, S.; Llambrich, A.; Zaballos, P.; et al.
Dermoscopic Clues for Diagnosing Melanomas That Resemble Seborrheic Keratosis. JAMA Dermatol. 2017, 153, 544–551. [CrossRef]
[PubMed]
27. Popadić, M.; Sinz, C.; Kittler, H. The Significance of Blue Color in Dermatoscopy. JDDG J. Der Dtsch. Dermatol. Ges. 2017,
15, 302–307. [CrossRef] [PubMed]
28. Manci, R.N.; Dauscher, M.; Marchetti, M.A.; Usatine, R.; Rotemberg, V.; Dusza, S.W.; Marghoob, A.A. Features of Skin Cancer
in Black Individuals: A Single-Institution Retrospective Cohort Study. Dermatol. Pract. Concept. 2022, 12, e2022075. [CrossRef]
[PubMed]
29. Golińska, J.; Sar-Pomian, M.; Rudnicka, L. Dermoscopic Features of Psoriasis of the Skin, Scalp and Nails—A Systematic Review.
J. Eur. Acad. Dermatol. Venereol. 2019, 33, 648–660. [CrossRef]
Diagnostics 2024, 14, 1165 21 of 21

30. Liu, M.; Chen, H.; Xu, F. Dermoscopy of Cutaneous Sarcoidosis: A Cross-Sectional Study. An. Bras. Dermatol. 2023, 98, 750–754.
[CrossRef]
31. Das, A.; Madke, B.; Jakhar, D.; Neema, S.; Kaur, I.; Kumar, P.; Pradhan, S. Named Signs and Metaphoric Terminologies in
Dermoscopy: A Compilation. Indian. J. Dermatol. Venereol. Leprol. 2022, 88, 855. [CrossRef] [PubMed]
32. Kandwal, M.; Jindal, R.; Chauhan, P.; Roy, S. Skin Diseases in Geriatrics and Their Effect on the Quality of Life: A Hospital-Based
Observational Study. J. Fam. Med. Prim. Care 2020, 9, 1453. [CrossRef] [PubMed]
33. Lallas, A.; Apalla, Z.; Ioannides, D.; Lazaridou, E.; Kyrgidis, A.; Broganelli, P.; Alfano, R.; Zalaudek, I.; Argenziano, G.;
Bakos, R.; et al. Update on Dermoscopy of Spitz/Reed Naevi and Management Guidelines by the International Dermoscopy
Society. Br. J. Dermatol. 2017, 177, 645–655. [CrossRef] [PubMed]
34. Enechukwu, N.A.; Behera, B.; Ding, D.D.; Lallas, A.; Chauhan, P.; Khare, S.; Sławińska, M.; Nisa Akay, B.; Ankad, B.S.;
Bhat, Y.J.; et al. Dermoscopy of Cutaneous Neoplasms in Skin of Color—A Systematic Review by the International Dermoscopy
Society “Imaging in Skin of Color” Task Force. Dermatol. Pract. Concept. 2023, 13, e2023308S. [CrossRef] [PubMed]
35. Griffith, C.F. Skin Cancer in Immunosuppressed Patients. JAAPA 2022, 35, 19–27. [CrossRef] [PubMed]
36. Singh, R.; Jawade, S.; Madke, B. Actinic Lichen Planus: Significance of Dermoscopic Assessment. Cureus 2023, 15, e35716.
[CrossRef] [PubMed]
37. Arpaia, N.; Cassano, N.; Vena, G.A. Dermoscopic Patterns of Dermatofibroma. Dermatol. Surg. 2006, 31, 1336–1339. [CrossRef]
[PubMed]
38. Papageorgiou, C.; Apalla, Z.; Variaah, G.; Matiaki, F.C.; Sotiriou, E.; Vakirlis, E.; Lazaridou, E.; Ioannides, D.; Lallas, A. Accuracy
of Dermoscopic Criteria for the Differentiation between Superficial Basal Cell Carcinoma and Bowen’s Disease. J. Eur. Acad.
Dermatol. Venereol. 2018, 32, 1914–1919. [CrossRef] [PubMed]
39. Karampinis, E.; Lallas, A.; Lazaridou, E.; Errichetti, E.; Apalla, Z. Race-Specific and Skin of Color Dermatoscopic Characteristics
of Skin Cancer: A Literature Review. Dermatol. Pract. Concept. 2023, 13, e2023311S. [CrossRef]
40. Bang, K.M.; Halder, R.M.; White, J.E.; Sampson, C.C.; Wilson, J. Skin Cancer in Black Americans: A Review of 126 Cases. J. Natl.
Med. Assoc. 1987, 79, 51–58.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.