RadiomicNet: A Hybrid Radiomics-Guided Lightweight Architecture for Interpretable Medical Image Segmentation

2026-07-02 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, quick

Summary

RadiomicNet is a novel two-stream hybrid architecture designed for interpretable medical image segmentation, addressing deep learning's limitations in tractability, parameter requirements, and clinical interpretability. It integrates handcrafted radiomics features, specifically Gray-Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP), directly into a lightweight MobileNetV2-based encoder-decoder via a Radiomics Attention Gate (RAG). A Radiomics Consistency Loss further aligns texture complexity with prediction uncertainty, reducing Expected Calibration Error (ECE) from 0.142 to 0.118. RadiomicNet achieves a Dice Similarity Coefficient (DSC) of 0.763 +/- 0.231 on the Breast Ultrasound Images (BUSI) dataset and 0.854 +/- 0.112 on Kvasir-SEG, outperforming U-KAN by 1.2% and 1.8%. It operates with only 3.27M parameters, 9.5x fewer than standard U-Net.

Key takeaway

For Machine Learning Engineers developing medical image segmentation models, RadiomicNet offers a path to achieve high performance with significantly reduced computational overhead and built-in interpretability. If your project requires compact models or clinical explainability, consider integrating handcrafted radiomics features via attention mechanisms. This approach can yield models with 9.5x fewer parameters than U-Net while improving Dice Similarity Coefficient and reducing Expected Calibration Error.

Key insights

RadiomicNet integrates handcrafted radiomics features into deep learning for interpretable, lightweight medical image segmentation.

Principles

• Hybrid architectures can enhance interpretability.
• Domain knowledge improves model efficiency.
• Attention mechanisms can integrate feature types.

Method

RadiomicNet uses a two-stream MobileNetV2-based encoder-decoder with a Radiomics Attention Gate (RAG) for GLCM/LBP feature integration and a Radiomics Consistency Loss.

In practice

• Utilize GLCM and LBP for texture analysis.
• Employ RAG for skip-connection modulation.
• Use consistency loss to reduce ECE.

Topics

• Medical Image Segmentation
• Radiomics
• Deep Learning
• Model Interpretability
• Attention Mechanisms
• Lightweight Architectures

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

Related on AIssential

• Attention-Gated Convolutional Networks for Scanner-Agnostic Quality Assessment — A hybrid CNN-Attention model effectively assesses MRI quality, generalizing across diverse scanners without retraining.
• Improving Pre-trained Adult Glioma Segmentation Models Using only Post-processing Techniques — Adaptive, radiomic-based post-processing significantly enhances pre-trained glioma segmentation model accuracy without additional GPU resources.
• SegGuidedNet: Sub-Region-Aware Attention Supervision for Interpretable Brain Tumor Segmentation — SegGuidedNet achieves accurate brain tumor sub-region segmentation with built-in interpretability through sub-region-aware attention supervision.
• DSU-Net: An Attention-Enhanced Dense Skip U-Net for Breast Lesion Segmentation in Mammographic Images — DSU-Net combines dense skip connections and attention for accurate breast lesion segmentation in mammography.
• Scalable Training of Spatially Grounded 2D Vision-Language Models for Radiology — Spatially grounded radiology VLMs can be trained at scale without manual annotations, maintaining language quality.
• GloResNet: A lightweight 3D CNN with global topological features for preterm brain injury prediction — Lightweight 3D CNN with global topological features predicts preterm brain injury effectively.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.