Related Experiment Video
Updated: May 13, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
LiteFeatNet: A parameter-efficient and performance-centric deep learning model for multi-ocular disease
Usman Rafi1, Qamar Nawaz1, Muhammad Ahsan Latif1
1Department of Computer Science, Faculty of Sciences, University of Agriculture, Faisalabad, Punjab, Pakistan.
Plos One
|May 11, 2026
Summary
LiteFeatNet, a novel Convolutional Neural Network (CNN), efficiently identifies retinal conditions with fewer resources. This lightweight model achieves high accuracy and fast inference, making it ideal for low-resource clinical settings.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Convolutional Neural Networks (CNNs) demand substantial data and computational power for accurate retinal condition identification.
- Developing and deploying these models on resource-constrained devices presents significant challenges.
Purpose of the Study:
- To introduce LiteFeatNet, a robust and computationally efficient CNN for retinal condition prediction.
- To reduce the number of trainable parameters, computational resources, and processing time required for accurate identification.
Main Methods:
- Utilized a pre-trained NASNetMobile backbone for enhanced robustness and reduced computation.
- Proposed a time-efficient method for discriminative feature extraction from deep intermediate layers.
- Implemented a spatially-aware feature map reduction module and a custom classification module.
Main Results:
- LiteFeatNet achieved a testing accuracy of 90.33% on the Retinal Fundus Multi-Disease Image Dataset (RFMiD).
- Demonstrated a fast inference time of 4 milliseconds per image, outperforming twelve state-of-the-art models.
- Validated generalizability, scalability, and computational efficiency on external datasets and increased class complexity.
Conclusions:
- LiteFeatNet is a lightweight, fast, and robust CNN suitable for deployment in low-resource clinical settings.
- The synergistic combination of deep feature extraction and feature map refinement is key to its success.