Hybrid Approach for Enhancing Lesion Segmentation in Fundus Images

2026-04-16 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research, Medical Devices & Health Technology · Depth: Expert, extended

Summary

A novel hybrid model has been developed to enhance the segmentation of choroidal nevi (CN) lesions in fundus images, addressing challenges like indistinct borders, noise sensitivity, and dependency on large, high-quality datasets. This approach combines traditional mathematical/clustering segmentation methods, specifically SLIC, with insights from a U-Net convolutional neural network. The hybrid model achieved a Dice coefficient of 89.7% and an IoU of 80.01% on 1024x1024 fundus images, significantly outperforming the Attention U-Net model, which scored 51.3% and 34.2% respectively. It also demonstrated superior generalizability on external datasets and operates efficiently on standard CPUs, making it suitable for resource-constrained clinical environments. This work is part of a broader effort to create a decision support system for early CN diagnosis and monitoring.

Key takeaway

For AI Scientists and Computer Vision Engineers developing medical image diagnostic tools, this hybrid segmentation approach offers a compelling solution for choroidal nevus detection. Your teams should consider integrating traditional segmentation methods with lightweight U-Net models to achieve high accuracy on large, high-resolution images while significantly reducing GPU dependency and improving generalizability. This strategy can lead to more practical, deployable solutions for clinical settings, especially where hardware resources are limited.

Key insights

A hybrid model combining U-Net and SLIC significantly improves choroidal nevus segmentation accuracy and generalizability on fundus images.

Principles

• Smaller image sizes improve U-Net segmentation accuracy.
• Hybrid models can reduce dataset dependency and computational cost.
• Traditional segmentation methods benefit from automated parameter selection.

Method

The proposed method trains a small-sized U-Net (128x128) to generate parameters for a traditional SLIC segmentation model, which then segments full-size (1024x1024 or 3900x3900) images, automating parameter selection and improving efficiency.

In practice

• Use 128x128 image sizes for U-Net training to optimize accuracy.
• Integrate traditional segmentation with ML for resource-efficient inference.
• Automate SLIC parameter selection using U-Net outputs.

Topics

• Choroidal Nevi Segmentation
• Hybrid Segmentation Model
• U-Net Architecture
• SLIC Algorithm
• Fundus Image Analysis

Best for: AI Scientist, Research Scientist, Computer Vision Engineer

Related on AIssential

• Efficient Transformer-Based Localized Patch Sampling for Choroid Plexus Segmentation in Multiple Sclerosis — A SwinUNETR pipeline with localized patch sampling accurately segments choroid plexus in MS with 99% less computational cost.
• HQF-Net: A Hybrid Quantum-Classical Multi-Scale Fusion Network for Remote Sensing Image Segmentation — Hybrid quantum-classical networks can significantly improve remote sensing semantic segmentation by fusing multi-scale features and quantum-enhanced refinement.
• Prodigy-Segment for Pixel Segmentation — AI-assisted bounding box annotation significantly streamlines pixel-level segmentation tasks.
• Diffusion Autoencoder for Unsupervised Artifact Restoration in Handheld Fundus Images — An unsupervised diffusion autoencoder restores fundus image artifacts by learning semantic representations from high-quality data.
• SegMix:Shuffle-based Feedback Learning for Semantic Segmentation of Pathology Images — SegMix uses shuffle-based feedback learning to generate high-quality pseudo-segmentation masks from image-level labels.
• U-Net Explained: The CNN That Revolutionized Image Segmentation — U-Net's U-shaped encoder-decoder with skip connections enables precise pixel-level segmentation from limited data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.