Glaucoma Detection: The Optic-Guard: Dr. Ashok Kumar

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Glaucoma Detection:

The Optic-Guard
Guide : Dr. Ashok Kumar
Presented By:
Puneeth S [1AH20CS075]
Rakshitha K [1AH20CS086]
Ritika P [1AH20CS089]
Shree Harsha K V [1AH20CS099]
Problem Statement

Solution

Objectives

Methodology
AGENDA

Code, Screenshots & Result

Future Enhancement

References
Problem Statement

➢ Glaucoma, a progressive optic neuropathy, is a major cause of


irreversible blindness globally. Timely detection is crucial for
effective intervention and preservation of vision. Current
diagnostic methods often rely on clinical assessments, and
there is a pressing need for a robust, automated system that
utilizes machine learning to enhance early detection.

➢ Existing glaucoma detection methods face challenges in terms


of accuracy, efficiency, and scalability. Manual assessment is
time-consuming, prone to subjectivity, and may miss early
signs. Leveraging advancements in machine learning, this
project aims to develop an automated glaucoma detection
system using retinal images.
Problem Statement Solution

➢ The Optic Guard project aims to address the problem of early detection of
glaucoma using innovative technological solutions. Glaucoma is a serious eye
condition that can lead to irreversible vision loss if not detected and treated
early. The project's solution involves the development of a system that utilizes
advanced imaging technology and artificial intelligence algorithms to detect
signs of glaucoma in its early stages.

➢ Imaging Technology: The project utilizes high-resolution imaging technology


such as Optical Coherence Tomography (OCT) or fundus photography to
capture detailed images of the optic nerve and the retina. These images provide
valuable information about the structural changes associated with glaucoma.

➢ Artificial Intelligence Algorithms: The extracted data is fed into advanced


artificial intelligence algorithms, such as deep learning neural networks, which
have been trained on a large dataset of images from both healthy individuals
and those with diagnosed glaucoma. These algorithms can effectively identify
patterns and abnormalities indicative of glaucoma with high accuracy.
METHODOLOGY

Disease Analysis ➢ Input of


human
Number of cases of the disease retina
over time
● Graph shows the cases trend of few regions.
● These cases are increasing rapidly over
time.

➢ Output

Region Primary Stage Secondary


Stage

North 5,306 419

South 3,543 253

East 4,654 342


Diagnosis & treatment flowchart

Start

Input the test data Convert to Grey Image Processing Detection

The captured images are resized The pixeled images are Through IP the cup-to-disk ratio In the detection stage if declared
to the required pixels and then converted to grey filters to have is calculated by the algorithm. that the patient is affected from
fed into the algorithm a accurate prediction. glaucoma the relevant
information will be displayed on
the monitor

And client will be notified.

Patient
Notification
Modules

1. Pre-processing module

Image Acquisition: The process begins with acquiring the images of interest, which
could be retinal images in the case of glaucoma detection or any other type of images
relevant to the analysis.

Image Enhancement: Preprocessing may involve image enhancement techniques to


improve the quality and clarity of the images. This could include methods such as
contrast stretching, histogram equalization, or filtering to enhance features of interest
and reduce noise.

Normalization: In some cases, it may be necessary to normalize the images to ensure


consistency in intensity values across different images. Normalization techniques adjust
the intensity range of pixels to a standard scale, facilitating more reliable clustering
results.
2. Optic-nerve segmentation module:

Density-based Thresholding: This method involves setting intensity thresholds to


distinguish the optic nerve head from surrounding retinal tissue based on pixel intensity
values. Adaptive thresholding techniques may be employed to account for variations in
illumination and contrast.

Edge Detection: Edge detection algorithms, such as Canny edge detector or Sobel
operator, can be used to identify edges or boundaries of the optic nerve head. Detected
edges can then be refined and connected to form the segmented region.

Region Growing: Region growing algorithms start from seed points within the optic
nerve head region and iteratively expand the region based on similarity criteria, such as
pixel intensity or texture features.
Glaucoma diognisis diagram
Code
1. K-means clustering algorithm

K=2
ret, label1, center1 = cv2.kmeans(Z, K, None, criteria, 10,
cv2.KMEANS_RANDOM_CENTERS)
center1 = np.uint8(center1)
res1 = center1[label1.flatten()]
output1 = res1.reshape((lab.shape))

bilateral_filtered_image = cv2.bilateralFilter(output1, 5, 175, 175)


edge_detected_image = cv2.Canny(bilateral_filtered_image, 75, 200)

contours, _= cv2.findContours(edge_detected_image,
cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contour_list = []
for contour in contours:
approx =
cv2.approxPolyDP(contour,0.01*cv2.arcLength(contour,True),True)
area = cv2.contourArea(contour)
if ((len(approx) > 8) & (area > 30) ):
contour_list.append(contour)
cv2.drawContours(image, contour_list, -1, (255,0,0), 1)K=2
ret, label1, center1 = cv2.kmeans(Z, K, None, criteria, 10,
cv2.KMEANS_RANDOM_CENTERS)
center1 = np.uint8(center1)
res1 = center1[label1.flatten()]
output1 = res1.reshape((lab.shape))
image = cv2.resize(image, (256, 256), CDR Process
interpolation=cv2.INTER_AREA)
HSV_img = cv2.cvtColor(merge,cv2.COLOR_RGB2HSV)
orig = image.copy() h,s,v = cv2.split(HSV_img)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (3, 3), 0) median = cv2.medianBlur(s,5)
(minVal, maxVal, minLoc, maxLoc) = cv2.minMaxLoc(gray) merge1 = cv2.merge((h,s,median))
cv2.circle(image, maxLoc, 80, (0, 0, 0), 2)
disc = 3.14 * 80 * 80
print('Area of Disc:'+str(disc))

r,g,b = cv2.split(orig)
kernel = np.ones((5,5), np.uint8)
img_dilation = cv2.dilate(g, kernel, iterations=1)

minmax_img =
np.zeros((img_dilation.shape[0],img_dilation.shape[1]),dtype =
'uint8')

for i in range(img_dilation.shape[0]):
for j in range(img_dilation.shape[1]):
minmax_img[i,j] = 255*(img_dilation[i,j]-
np.min(img_dilation))/(np.max(img_dilation)-
np.min(img_dilation))

merge = cv2.merge((r,minmax_img,b))
def display(image,panel):
GUI Code root.title(“Optic-Guard”)
root.configure(background='#4798FF')
image =Image.fromarray(image)
root.geometry("1200x500")
image=ImageTk.PhotoImage(image)
s=ttk.Style()
if panel is None:
s.configure('TButton',
panel=Label(image=image)
font=('calibri',15),foreground='black',anchor=CENTER,borde
panel.image=image
rwidth=2,background='white')
panel.pack(side="left",padx=10,pady=10)
text=Text(root,height=1,
else:
width='34',font=('Serif',15,'bold'),foreground='black',bor
panel.configure(image=image)
derwidth=2,background='white')
panel.image=image
text.insert(INSERT,"\tGlaucoma Detection\n")
text.pack()
def select_image():
btnml = ttk.Button(text="Machine Learning + IP“ ,command =
global panelA, panelB,prashu
ml)
path = filedialog.askopenfilename()
btnml.pack(side="bottom", expand="no", padx="10",
if len(path) > 0:
pady="10")
image = cv2.imread(path)
btnip = ttk.Button(text = "Image Processing", command = ip
prashu=image.copy()
)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
btnip.pack(side="bottom", expand="no", padx="10",
image = Image.fromarray(image)
pady="10")
image = ImageTk.PhotoImage(image)
btn = Button(root, text="Select an image",
if panelA is None or panelB is None:
command=select_image)
panelA = Label(image=image)
btn.pack(side="bottom",expand="no", padx="10", pady="10")
panelA.image = image
root.mainloop()
panelA.pack(side="left", padx=10, pady=10)
else:
panelA.configure(image=image)
panelA.image = image
Screenshots
Twilio API Workflow
Result

The results of a glaucoma detection project can vary depending on the specific objectives,
methodologies, and datasets used. Here are some key results typically reported in such projects:

• Diagnostic Accuracy Metrics:

These include sensitivity, specificity, positive predictive value (PPV), negative predictive value
(NPV), and area under the receiver operating characteristic curve (AUC-ROC). These metrics
quantify the performance of the developed model or system in distinguishing between patients
with glaucoma and those without.

• Comparison with Existing Methods:

Results may include comparisons of the developed approach with existing diagnostic methods,
such as clinical assessment by ophthalmologists, visual field tests, or other imaging modalities.
Demonstrating superior performance or complementary capabilities can validate the effectiveness
of the developed approach.
Future Enhancements

Future enhancements for glaucoma detection projects may involve leveraging emerging
technologies, refining existing methodologies, and addressing current limitations. Here are some
potential avenues for enhancement:

1. Advanced Machine Learning Techniques:


Exploring advanced ML techniques, such as deep learning architectures (e.g., convolutional
neural networks, recurrent neural networks), reinforcement learning, and unsupervised learning,
can further enhance the performance of glaucoma detection algorithms. These techniques can
learn complex patterns and relationships from large datasets, leading to improved predictive
models and feature representations.

2. Personalized Risk Assessment:


Moving towards personalized medicine, future enhancements may involve developing algorithms
capable of individualized risk assessment for glaucoma. By considering patient-specific factors,
such as demographics, genetics, medical history, and lifestyle factors, tailored risk prediction
models can provide more accurate assessments of an individual's likelihood of developing
glaucoma and guide personalized interventions.
References

I. References Gu, B., Sidhu, S., Weinreb, R. N., & Cheung, C. Y. (2022). Review of visualization approaches in
deep learning models of glaucoma. The Asia-Pacific Journal of Ophthalmology, 11(2), 121-132.

II. Christopher, M., Nakahara, K., Bowd, C., Medeiros, F. A., Weinreb, R. N., & Cheung, C. Y. (2020). Effects of
study population, labeling and training on glaucoma detection using deep learning algorithms. Translational
Vision Science & Technology, 9(5), 43.

III. Zhang, X., Li, F., Wang, D., & Lam, D. S. C. (2023). Visualization Techniques to Enhance the Explainability
and Usability of Deep Learning Models in Glaucoma. The Asia-Pacific Journal of Ophthalmology, 12(2), 146-157.

IV. Sharmila, C., & Shanthi, N. (2021). An exploration on deep learning approaches for the detection of
glaucoma. 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS) (pp. 1396-
1401). IEEE.

V. García, G., Colomer, A., & Naranjo, V. (2021). Glaucoma detection from raw SD-OCT volumes: a novel
approach focused on spatial dependencies. Computer Methods and Programs in Biomedicine, 202, 105956.

VI. Mayro, E. L., Wang, M., Elze, T., & Pasquale, L. R. (2020). The impact of artificial intelligence in the diagnosis
and management of glaucoma. Eye, 34(10), 1748-1763.
THANK-YOU

Thankyou.

You might also like