PORTABLE

CATARACT
DETECTION
AND GRADING
SYSTEM
 INDEX
Sr. No. Title Page No.

1 Block Diagram

2 Dataset

3 Results of MobileNet CNN

4 Conclusion

5 YOLOv8

6 Hough Transform

7 Hardware Requirements

8 Basic App Design

9 MongoDB Atlas
Block Diagram
Consultant Ophthalmologist at Manipal Hospital, Goa.
DR. KAUSHIK DHUME

QUALIFICATION:
MBBS | MD Physician | DNB
(Ophthalmology)

Field of Expertise
● Cataract Surgeries
● Glaucoma
● Anterior Segment Surgery
● Lasers

21 Years Experience Overall

(12 years as specialist)
 Improved
Dataset

Cortical Mature Cortical

NS1

NS2 NS3
PSC
 Total = 5722
Training Dataset
Classes Count of images
Cortical 800
Mature Cortical 854
NS Grade 1 811
NS Grade 2 804
NS Grade 3 805
Posterior Subcapsular 804
Normal 844
 Testing Dataset TOTAL = 1395

Classes Count of images

Cortical 200
Mature Cortical 198
NS Grade 1 201
NS Grade 2 200
NS Grade 3 198
Posterior Subcapsular 200
Normal 198
 Dataset size

Total Dataset size = 7117

 Results of MobileNet CNN for 12 Epochs and 32 Batch size

Test Accuracy: 0.5627

Precision: 0.5983
Recall: 0.5627
F1 Score: 0.5332
Overall Specificity 0.6907
Results of MobileNet CNN for 12 Epochs and 32 Batch size
 Results of MobileNet CNN for 25 Epochs and 64 Batch size

Test Accuracy: 0.5369

Precision: 0.6042
Recall: 0.5369
F1 Score: 0.5101
Overall Specificity 0.6307
Results of MobileNet CNN for 25 Epochs and 64 Batch size
 Results of MobileNet CNN for 35 Epochs and 128 Batch
size

Test Accuracy: 0.5297

Precision: 0.5773
Recall: 0.5297
F1 Score: 0.4968
Overall Specificity 0.6541
Results of MobileNet CNN for 35 Epochs and 128 Batch
size
 Conclusion

12 Epochs & 32 25 Epochs & 64 35 Epochs &

Batch size Batch size 128 Batch size

Test Accuracy: 0.5627 0.5369 0.5297

Precision: 0.5983 0.6042 0.5773

Recall: 0.5627 0.5369 0.5297

F1 Score: 0.5332 0.5101 0.4968

Overall 0.6907 0.6307 0.6541

Specificity
 YOLOv8
Used for Iris Detection

● Latest version in the YOLO series of real-time object detectors.

● Offers cutting-edge performance in terms of accuracy and speed.
● Builds upon the advancements of previous YOLO versions.

Tasks
● Image Classification : involves categorizing an entire image without localizing
the object present within the image.
● Object Detection : localizes an object within an image by drawing bounding
boxes.
● Image Segmentation : identifies each pixel belonging to an object and is more
precise in locating different objects within a single image.
Models Available in
YOLOv8
● There are six models in each category of YOLOv8 models for detection,
segmentation, and classification.

● YOLOv8 Nano (YOLOv8n) is the fastest and smallest, while YOLOv8 Extra Large
(YOLOv8x) is the most accurate yet the slowest among them.
YOLOv8 Nano (YOLOv8n)
● "v8" in "YOLOv8" signifies the version of the YOLO architecture and each YOLO version
brings improvements, modifications, or enhancements over the previous version.
● "Nano" indicates a lightweight, compact model and designed for resource-constrained
devices like mobile phones, embedded systems, or IoT devices.
● Models labeled as "Nano" often prioritize speed and efficiency over maximum accuracy
and may sacrifice some performance to achieve a smaller model size and faster
inference speed.

Key Features
● Lightweight Architecture : The YOLOv8 Nano model likely has a simplified or compressed
architecture compared to larger YOLO models to reduce computational and memory
requirements.
● Efficient Inference : It's designed to perform inference quickly, making it suitable for real-
time applications with limited computational resources.
● Deployment : YOLOv8 Nano models are often intended for deployment in scenarios where
real-time object detection is crucial, such as in robotics, autonomous vehicles, or smart
surveillance systems.
YOLO Annotation (Dataset obtained from : Roboflow universe)
● The annotation format for YOLO typically involves creating a text file for each image in the dataset.
● Each text file contains a line for each object in the image, specifying the object's class label and its
bounding box coordinates relative to the image size.

<object-class> <x-center> <y-center> <width> <height>

Where :

● <object-class>: The category or type of the detected object, usually given as a number.
● <x-center>: The horizontal position of the center of the bounding box, scaled to the image width.
● <y-center>: The vertical position of the center of the bounding box, scaled to the image height.
● <width>: The width of the bounding box, scaled to the image width.
● <height>: The height of the bounding box, scaled to the image height.

Example : 0 0.540022 0.678142 0.266652 0.244262

This indicates that an object of class 0 (or the first class in the dataset) has been detected with its
bounding box center at (0.540022, 0.678142) and a width of 0.266652 and height of 0.244262.
What is Hough Transform?
The Hough Transform is a technique used in computer vision and image
processing to detect shapes, primarily lines and curves, within an image.

The basic idea behind the Hough Transform is to identify shapes in an

image by transforming the image space into a parameter space, where
each point in the image corresponds to a curve or line in the parameter
space. This transformation allows for the detection of shapes even in the
presence of noise, gaps, or other imperfections in the image.

Iris Tracking: Pupil Tracking:

Hough Circle Transform:

The Hough Circle Transform is a technique used in computer vision and image
processing to detect circles within an image.

Working of Hough Circle Transform :

1. Edge Detection
2. Parameter Space Representation
3. Voting Process
4. Thresholding and Circle Detection
5. Circle Extraction
 Hardware Requirements
Sr. No. HARDWARE COST

1 Raspberry Pi 4 Model B ₹7000

2 Battery ₹600

3 HDMI Port / Pi Zero ₹1000/₹3000

4 Camera ₹500x2

5 Segmented LCD ₹1050

6 LED ₹95(each)

7 Case ₹1250
 Hardware Requirements
Sr. No. HARDWARE COST

8 Buck ₹250x2

9 Switch ₹30

10 Relay ₹50

11 15V Charger ₹600

12 Multi Camera Module 5800

APPROXIMATE TOTAL COST ₹192070

Fig 1.1: Raspberry Pi 4 Model B
 Fig 3: Raspberry Pi 4 Model B
Fig 2: Raspberry Pi 4 Model B HDMI
Battery
Fig 4: Raspberry Pi 4 Model B Fig 5: Raspberry Pi 4 Model B
Camera Segmented Display
Fig 6: Raspberry Pi 4 Model B Fig 7: Raspberry Pi 4 Model B
LED (Yellow) Buck Converter
 Fig 9: Raspberry Pi 4 Model B
Fig 8: Raspberry Pi 4 Model B Mosfet
Switch
Fig 10: Raspberry Pi 4 Model B Fig 11: Raspberry Pi 4 Model B
Relay 15V Charger
Fig 12: AUTOCAD 3D Model
Representation of Portable
Cataract Detection and Grading
System
BASIC APP
DESIGN:
Uploading and fetching of image
through html
Mongodb Atlas
Cloudinary : cloud storage