IDF-Net - Interpretable Dynamic Fusion Network for Colorectal Cancer Diagnosis Using Cross-Modal Imaging

Q: What problem does IDF-Net target in colorectal cancer diagnosis?

IDF-Net targets the need for early, accurate, clinically interpretable CRC diagnosis. It addresses shortcomings of unimodal AI and insufficient interpretability in existing models.

Q: How does IDF-Net fuse multiple imaging modalities?

IDF-Net integrates endoscopy, CT, and histopathology using modality-specific encoders, a dual-stage adaptive gating mechanism, and cross-modal attention. This design adaptively weights modalities based on patient-specific diagnostic relevance.

Q: What evidence supports the model’s interpretability and performance?

Interpretability is assessed with spatial heatmaps and modality attribution, including Gradient-weighted Class Activation Mapping++ alignment and SHapley Additive exPlanations values. Results report state-of-the-art accuracy and AUC, outperforming unimodal and static-fusion baselines.

Sep 30, 2026

Uploaded by Aria Callaghan

Colorectal cancer (CRC) remains a leading cause of cancer mortality, creating a strong need for diagnostic tools that are early, accurate, and clinically interpretable. Many current AI models are unimodal and provide limited interpretability, restricting clinical adoption. IDF-Net addresses this with an interpretable dynamic fusion framework combining endoscopy, CT, and histopathology via modality-specific encoders, dual-stage adaptive gating, and cross-modal attention. Evaluations using stratified 5-fold cross-validation show state-of-the-art performance and quantitative interpretability aligned with expert lesions, with robust operation even when CT is absent.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Introduction

Background and clinical motivation

Limitations of current AI and multimodal fusion

Proposed solution and contributions

Article IDF-Net: Interpretable Dynamic Fusion Network for Colorectal Cancer Diagnosis Using Cross-Modal Imaging Helen Haile Hayeso , Peifeng Shi , Jingwen Lian, Zenebe Markos Lonseko and Nini Rao * Department of Biomedical Engineering, University of Electronic Science and Technology of China, Chengdu 610054, China; [helenh@std.uestc.edu.cn](helenh@std.uestc.edu.cn) (H.H.H.); [202421140102@std.uestc.edu.cn](202421140102@std.uestc.edu.cn) (P.S.); [lianjingwen_xoxo@163.com](lianjingwen_xoxo@163.com) (J.L.); [zenebe@uestc.edu.cn](zenebe@uestc.edu.cn) (Z.M.L.) * Correspondence: [raonn@uestc.edu.cn](raonn@uestc.edu.cn) Abstract Background/Objectives: Colorectal cancer (CRC) is a leading cause of cancer deaths worldwide, underscoring the need for diagnostic tools that early, accurate, and clinically interpretable. Current artificial intelligence (AI) models are predominantly unimodal and lack sufficient interpretability, which restricts their clinical adoption. Methods: We propose IDF-Net, an interpretable dynamic fusion framework that integrates endoscopy, computed tomography (CT), and histopathology using modality-specific encoders, a dual-stage adaptive gating mechanism, and cross-modal attention. We conducted stratified 5-fold cross-validation and assessed interpretability using spatial heatmaps and modality attribution. We also quantified the results using the intersection-over-union metric for saliency alignment. Results: IDF-Net achieved a state-of-the-art accuracy of 0.920 (0.907–0.936) and area under the curve (AUC) of 0.991 (95% CI: 0.965–0.997), significantly outperforming unimodal and static-fusion baselines (p < 0.05) . Interpretability analysis of IDF-Net demonstrated a strong alignment between Gradient-weighted Class Activation Mapping++ heatmaps and expert-annotated lesions, as well as case-specific modality contributions via SHapley Additive exPlanations values. Ablation studies confirmed the contribution of each component, with dynamic routing and cross-attention fusion improving AUC by 0.038 and 0.046, respectively. Conclusions: IDF-Net introduces a dynamically fused, multimodal diagnostic framework with integrated quantitative interpretability, demonstrating superior accuracy and strong potential for clinical translation in CRC diagnosis. The model’s adaptive design allows it to function robustly even when CT data is unavailable, aligning with common clinical pathways while leveraging additional imaging when present for comprehensive staging. Keywords: colorectal cancer diagnosis; gastrointestinal disease; dynamic fusion network; deep learning; machine learning; cross-modal imaging; interpretable AI Academic Editor: Kwang-Sig Lee Received: 3 December 2025 Revised: 17 December 2025 Accepted: 25 December 2025 Published: 27 December 2025 Copyright: © 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. 1. Introduction Among gastrointestinal diseases (GIDs), colorectal cancer (CRC) poses a particularly formidable global health challenge, ranking as the third most diagnosed cancer and the second leading cause of cancer-related mortality worldwide [1,2] . Recent studies indicate approximately 1.9 million new cases and 935,000 deaths annually, with projections suggesting a dramatic rise in incidence by 2040 [1–3] . This growing burden, coupled with substantial economic costs, underscores an urgent need for enhanced diagnostic strategies that improve early detection, accurate staging, and severity estimation [4,5] . The current diagnostic paradigm depends on a multimodal diagnosis, including endoscopy for direct visualization, computed tomography (CT) for staging, and histopathology for definitive confirmation. However, these modalities are typically interpreted in isolation, resulting in fragmented assessment, potential diagnostic delays, and limite

FAQs

What problem does IDF-Net target in colorectal cancer diagnosis?

IDF-Net targets the need for early, accurate, clinically interpretable CRC diagnosis. It addresses shortcomings of unimodal AI and insufficient interpretability in existing models.

How does IDF-Net fuse multiple imaging modalities?

IDF-Net integrates endoscopy, CT, and histopathology using modality-specific encoders, a dual-stage adaptive gating mechanism, and cross-modal attention. This design adaptively weights modalities based on patient-specific diagnostic relevance.

What evidence supports the model’s interpretability and performance?

Interpretability is assessed with spatial heatmaps and modality attribution, including Gradient-weighted Class Activation Mapping++ alignment and SHapley Additive exPlanations values. Results report state-of-the-art accuracy and AUC, outperforming unimodal and static-fusion baselines.

Sep 30, 2026

Uploaded by Aria Callaghan