Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Advancing Medical Image Segmentation with Mini-Net: A Lightweight Solution Tailored for Efficient Segmentation of Medical Images

Syed Javed, Tariq M. Khan, Abdul Qayyum, Hamid Alinejad-Rokny, Arcot Sowmya, Imran Razzak

TL;DR

This work tackles the need for real-time, resource-efficient medical image segmentation by introducing Mini-Net, a lightweight encoder-decoder network built around dual multi-residual (DMRes) blocks and Expand-Squeeze blocks. With only about 37.7k parameters, Mini-Net achieves competitive or state-of-the-art performance across diverse datasets (retinal vessels, skin lesions, and cell nuclei) while maintaining real-time inference potential on edge devices. An alpha-weighted loss combining Dice, Jaccard, and BCE losses is employed to balance foreground-background segmentation and boundary accuracy, with ablations showing the best results on ISIC-2018. The contributions offer a practical, robust solution for deploying accurate medical segmentation on devices with limited memory and compute, enabling broader clinical applicability.

Abstract

Accurate segmentation of anatomical structures and abnormalities in medical images is crucial for computer-aided diagnosis and analysis. While deep learning techniques excel at this task, their computational demands pose challenges. Additionally, some cutting-edge segmentation methods, though effective for general object segmentation, may not be optimised for medical images. To address these issues, we propose Mini-Net, a lightweight segmentation network specifically designed for medical images. With fewer than 38,000 parameters, Mini-Net efficiently captures both high- and low-frequency features, enabling real-time applications in various medical imaging scenarios. We evaluate Mini-Net on various datasets, including DRIVE, STARE, ISIC-2016, ISIC-2018, and MoNuSeg, demonstrating its robustness and good performance compared to state-of-the-art methods.

Advancing Medical Image Segmentation with Mini-Net: A Lightweight Solution Tailored for Efficient Segmentation of Medical Images

TL;DR

This work tackles the need for real-time, resource-efficient medical image segmentation by introducing Mini-Net, a lightweight encoder-decoder network built around dual multi-residual (DMRes) blocks and Expand-Squeeze blocks. With only about 37.7k parameters, Mini-Net achieves competitive or state-of-the-art performance across diverse datasets (retinal vessels, skin lesions, and cell nuclei) while maintaining real-time inference potential on edge devices. An alpha-weighted loss combining Dice, Jaccard, and BCE losses is employed to balance foreground-background segmentation and boundary accuracy, with ablations showing the best results on ISIC-2018. The contributions offer a practical, robust solution for deploying accurate medical segmentation on devices with limited memory and compute, enabling broader clinical applicability.

Abstract

Accurate segmentation of anatomical structures and abnormalities in medical images is crucial for computer-aided diagnosis and analysis. While deep learning techniques excel at this task, their computational demands pose challenges. Additionally, some cutting-edge segmentation methods, though effective for general object segmentation, may not be optimised for medical images. To address these issues, we propose Mini-Net, a lightweight segmentation network specifically designed for medical images. With fewer than 38,000 parameters, Mini-Net efficiently captures both high- and low-frequency features, enabling real-time applications in various medical imaging scenarios. We evaluate Mini-Net on various datasets, including DRIVE, STARE, ISIC-2016, ISIC-2018, and MoNuSeg, demonstrating its robustness and good performance compared to state-of-the-art methods.
Paper Structure (12 sections, 16 equations, 2 figures, 6 tables)

This paper contains 12 sections, 16 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Supervised Deep Learning based Techniques
  4. Semi-supervised Deep Learning based Techniques
  5. Lightweight Medical Image Segmentation Models
  6. Methodology
  7. Mini-Net Architecture
  8. Loss Function
  9. Experiments and Implementation Details
  10. Results and Discussion
  11. Ablation Study
  12. Conclusions

Figures (2)

  • Figure 1: Mini-Net Model Diagram
  • Figure 2: Qualitative results of Mini-Net on sample images from (a) MonuSeg, (b) CHASE, and (c) ISIC-2018 datasets. The columns from left to right in each block represent query image, ground truth mask, and the predicted mask by Mini-Net respectively. The green and black pixels are the correctly segmented foreground and background respectively while blue pixels are the false positives and the red ones are the false negative pixels.