Scribd
Upload
16 views1 page

DR Table

The document presents a lightweight deep learning approach for diabetic retinopathy (DR) detection and classification using a Convolutional Neural Network (CNN) and EfficientNetB3. The model achieved a 95% validation accuracy for detection and 84% for classification, outperforming several existing models. It highlights research gaps such as sensitivity issues, the need for real-time solutions, and challenges with imbalanced datasets.

Uploaded by

dipobi9608
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
16 views1 page

DR Table

The document presents a lightweight deep learning approach for diabetic retinopathy (DR) detection and classification using a Convolutional Neural Network (CNN) and EfficientNetB3. The model achieved a 95% validation accuracy for detection and 84% for classification, outperforming several existing models. It highlights research gaps such as sensitivity issues, the need for real-time solutions, and challenges with imbalanced datasets.

Uploaded by

dipobi9608
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Title Diabetic Retinopathy Detection and Categorizing Using a Lightweight Deep

Learning Approach
Authors D. Praneeth and N. Satheesh Kumar Department of Computer Science and
Engineering, Chaitanya (Deemed to be University), Telangana, India
Year of Publication 2024 (Published in ARPN Journal of Engineering and Applied Sciences, Vol. 19,
No. 13)
Methodology Used A lightweight Convolutional Neural Network (CNN) for DR detection with fewer
trainable parameters for resource-constrained environments. EfficientNetB3
used for DR classification with compound scaling (depth, width, resolution).
Preprocessing: Gaussian filtering, image resizing to 224×224. Evaluation
metrics: Accuracy, Precision, Recall, F1-Score, Confusion Matrix.
Dataset Used APTOS 2019 Blindness Detection dataset (Kaggle). Fundus images labeled into
five classes: 0 – No_DR 1 – Mild 2 – Moderate 3 – Severe 4 – Proliferate_DR
Parameters Considered - Image size: 224×224 pixels - Features: microaneurysms, hemorrhages,
exudates, neovascularization - Evaluation metrics: Training/Validation Accuracy,
Precision, Recall, F1-Score, Confusion Matrix - CNN with reduced weights for
detection - EfficientNetB3 with compound scaling for classification
Result Detection Model (CNN): 95% validation accuracy, outperforming CNN (69%),
VGG (76%), EfficientNetB0 (78%), MobileNet (85). Classification Model
(EfficientNetB3): 84% validation accuracy, higher than ResNet50 (56%),
DenseNet (73%), InceptionV3 (74). Class-wise performance: - No_DR: Precision
0.99, Recall 0.98 - Mild: Precision 0.62, Recall 0.67 - Proliferate_DR: Precision
0.61, Recall 0.47
Research Gap - Sensitivity and specificity issues limit clinical reliability - Need real-time,
cost-effective, accessible DR screening solutions - Imbalanced datasets hinder
severe case detection (Moderate vs Proliferate_DR misclassification) -
Vulnerable to overfitting/underfitting, requiring better generalization for real-world
deployment

You might also like