Professional Documents
Culture Documents
Phase 3b
Phase 3b
By-
Mentors: Akangshi Lahary(1MS15ML005)
Dr.Sriraam, HOD, Medical Electronics, RIT. Apoorva Maiya(1MS15ML011)
Dr.Sharath, Research Scientist, CIT, RIT. Nishchal Bhat M(1MS17ML405)
Dr. Prabha Ravi, Associate Professor, Medical Electronics,RIT Vasanth Patil(1MS17ML410)
• In 2018, it is estimated that 627,000 women died from breast cancer – that is
approximately 15% of all cancer deaths among women.
• Breast cancer has ranked number one cancer among Indian females as high as 25.8 per
100,000 women and mortality 12.7 per 100,000 women.
• As per the Indian Council of Medical Research (ICMR) India had 14 Lakh cancer
patients in 2016 and this is number has been increasing since.
• There is a need for an automated CAD tool to reduce the manual error in breast cancer
screening.
• In a small study published in The American Journal of Surgery, Thermography
accurately identified 58 out 60 breast malignancies among 92 patients; accuracy rate of
97 per cent was obtained.
• Conventional breast cancer techniques like mammography pose the risk of exposure to
X-ray radiation a research showed that every time a subject is exposed to X-rays the risk
of cancer is increased by 2 per cent.
• Ultrasound cannot be used as a standalone tool; it is often used with mammography to
determine the type of the cancer cells. It also misses the early signs like
microcalcification.
EXISTING MODALITIES AND COMPARISION
Mammography Infrared imaging Ultrasound
Uses x-rays Uses infrared sensors to detect heat High frequency sound waves are
emitted by the body bounced of the breast tissue and collect
as an echo to produce an image
Early detection method Early detection method used as an Early detection method. Used as an
adjunctive image test adjunctive image test
Can detect tumours in pre-invasive stage May provide the first signal that a Lower spatial resolution. Good at
problem is developing distinguishing solid masses from fluid
filled cysts.
2. Automatic 200 Backgroun Canny edge First order SVM, After using our
image thermogram d removal detection statistic ANN proposed SM, the
segmentation (90 normal , and then operator (mean, accuracy of SVM
Method for 110 abnormal) resizing. With variance, and MLP
breast cancer (public) Hough skewness, Reached
analysis using transform kurtosis) & 96.667%, and
thermography (GLCM) 96.071%,
(oct,2017) respectively.
S Pre- Segmentati Features Class Result
Sl. No Paper processing on extracted ificat
Database
title/pub. year ion
• • Maximum
and minimum temperature values may differ from each
thermogram
• Average of every maximum and minimum value from each thermogram is
calculated and stored in maximum value and minimum value.
• The new image is normalized using the below equation:
Normalized Image
normalizing rgb
FILTERING METHODS
Following filters are tried out on all the images and their SNR values are tabulated and
compared. The chart shows SNR values of individual images and their mean SNR is
tabulated. Gaussian filter is chosen for noise reduction because of its high SNR i.e.
54.857.
70
60
50
40
Median Filter
30
Gaussian Filter
20
Mean Filter
10
0
-10
-20
SEGMENTATION
Automated segmentation is carried out using polynomial curve fitting technique
ROI ROI
FEATURE EXTRACTION
1.7
1.6
1.5
Normal
1.4 Abnormal
1.3
1.2
Fractal dimension
LACUNARITY
• Lacunarity is derived using Gliding box technique which uses the boxes of
same size and number as calculated for Fractal Dimension.
• Lacunarity is calculated for each box and the mean Lacunarity is obtained by
averaging over all Lacunarities for an image
• Mean Lacunarity for normal images was found to be 1.0287 and 1.049 for
abnormal images.
1.07
1.06
1.05
1.04 Normal
Abnormal
1.03
1.02
1.01
Lacunarity
GLCM from LAW’S ENERGY MAPS
• 4 1-D Laws texture filters are used, each filter enhance certain underlying
microstructure features as mentioned in the brackets.
L5 = [+1 +4 6 +4 +1] (Level); E5 = [-1 -2 0 +2 +1] (Edge); S5 = [-1 0 2 0 -1] (Spot)
R5 = [+1 -4 6 -4 +1] (Ripple)
• Combinations of these 4 Filters yield 16 2-D filters. After eliminating mirror filters,
9 texture maps are obtained and applied on each image.
• 3 GLCM features are extracted from these images- (Contrast, Energy, Correlation)
0.7
0.6
0.5
0.4
0.3 Normal
Abnormal
0.2
0.1
0
Contrast Energy Correlation
-0.1
CLASSIFICATION and PERFORMANCE EVALUATION
FUTURE SCOPE
• Algorithm can be expanded to a larger database
• Segmentation technique can be made more precise using other automated
techniques.
• Different classifier can be used to suit a much larger database
• A classifier can be further developed to identify if the cancerous breast is benign or
malignant type.
REFERENCES
[1] Antony Morales-Cervantes1, Eleazar Samuel Kolosovas-Machuca2*, Edgar
Guevara2,5, Mireya Maruris Reducindo3, Alix Berenice Bello Hernández3, Manuel
Ramos García 4, Francisco Javier González2” An automated method for the
evaluation of breast cancer using infrared thermography “
[2] R. Ramya Devi1 & G. S. Anandhamala1” Analysis of Breast Thermograms Using
Asymmetry in Infra-Mammary Curves”. Journal of Medical Systems (2019)
43:146
[3] Dayakshini Sathish1 · Surekha Kamath1 · Keerthana Prasad2 · Rajagopal
Kadavigere3 · Roshan J. Martis4” Asymmetry analysis of breast thermograms using
automated segmentation and texture features”
[4] Rangaraj M. Rangayyan1,2 and Thanh M. Nguyen1” Fractal Analysis of Contours
of Breast Masses in Mammograms”
[5]Acharya UR, Ng EYK, Tan JH, Sree SV. Thermography based breast cancer
detection using texture features and support vector machine. J Med Syst. 2012;
36:1503-10.
[6] R. Ramya Devi1 & G. S. Anandhamala1” Analysis of Breast Thermograms Using
Asymmetry in Infra-Mammary Curves”. Journal of Medical Systems (2019) 43:146
THANK YOU
Individual contribution Special thanks to
1. Dr. B Venkatraman, Mrs. M Menaka
1. Vasanth Patil (1MS17ML410)- IGCAR, Kalpakkam Tamil Nadu
Literature Review, Data pre-
2. Ramaiah hospital
processing
3. Dr. Sriraam, HOD, Medical Electronics
2. Nishchal Bhat M
dept., RIT
(1MS17ML405)- Data pre-
processing, Classification 4. Dr. Sharath D, Research Scientist,
Center for Imaging Technologies, RIT
3. Akangshi Lahary
(1MS15ML005)- Segmentation, 5. Dr. Prabha Ravi, Associate Professor,
GUI Medical Electronics, RIT
4. Apoorva Maiya (1MS15ML011)- 6. Mrs. Kavya, Research Scholar, Center
Feature Extraction, Classification. for Imaging Technologies, RIT
7. Mrs. Usha, Research Scholar, Center for
Imaging Technologies, RIT