DiaGuard: An Explainable Multi-Modal Deep Learning Framework for Joint Diabetic Retinopathy and Foot Ulcer Risk Screening

Diabetes mellitus induces concurrent microvascular complications across multiple organ systems, most notably Diabetic Retinopathy (DR) in the ocular fundus and Diabetic Foot Ulcers (DFU) in the lower extremities. Despite their shared vascular pathology, automated computer-aided diagnosis systems typically evaluate retinal and plantar lesions in isolated silos and operate as opaque black-box classifiers, limiting clinical trust during primary triage. This paper presents DiaGuard, an end-to-end, explainable multi-modal screening framework that unifies five-stage DR grading and binary DFU detection within a single clinical decision-support portal. The ocular branch employs a fine-tuned ResNet-50 architecture, while the plantar dermal branch leverages an EfficientNet-V2-S backbone trained with class-weighted Focal Loss and photometric-safe data augmentation to preserve diagnostic erythema and ischemic tissue cues without color distortion. To ensure clinical transparency, Gradient-weighted Class Activation Mapping (Grad-CAM) hooks are attached to the final convolutional blocks of both networks to generate real-time spatial attention overlays alongside a max-weighted microvascular risk index. Evaluated on the held-out stratified validation fold (𝑛=211 scans from 𝑁=1,055 plantar skin images), the DFU branch achieves an Area Under the Receiver Operating Characteristic Curve (ROC-AUC) of 0.9985, a Macro F1-score of 0.9763, 98.06% sensitivity, and 99.07% specificity. Deployed via a decoupled FastAPI inference engine and responsive web interface, DiaGuard demonstrates a reproducible, low-latency architecture for comprehensive outpatient diabetic screening.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24419797
Primary Topic
Diabetic Foot Ulcer Assessment and Management
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

DiaGuard: An Explainable Multi-Modal Deep Learning Framework for Joint Diabetic Retinopathy and Foot Ulcer Risk Screening

Nirali
Zenodo (CERN European Organization for Nuclear Research)
Diabetic Foot Ulcer Assessment and Management
preprint

DiaGuard: An Explainable Multi-Modal Deep Learning Framework for Joint Diabetic Retinopathy and Foot Ulcer Risk Screening

Nirali
preprint en

Abstract

Diabetes mellitus induces concurrent microvascular complications across multiple organ systems, most notably Diabetic Retinopathy (DR) in the ocular fundus and Diabetic Foot Ulcers (DFU) in the lower extremities. Despite their shared vascular pathology, automated computer-aided diagnosis systems typically evaluate retinal and plantar lesions in isolated silos and operate as opaque black-box classifiers, limiting clinical trust during primary triage. This paper presents DiaGuard, an end-to-end, explainable multi-modal screening framework that unifies five-stage DR grading and binary DFU detection within a single clinical decision-support portal. The ocular branch employs a fine-tuned ResNet-50 architecture, while the plantar dermal branch leverages an EfficientNet-V2-S backbone trained with class-weighted Focal Loss and photometric-safe data augmentation to preserve diagnostic erythema and ischemic tissue cues without color distortion. To ensure clinical transparency, Gradient-weighted Class Activation Mapping (Grad-CAM) hooks are attached to the final convolutional blocks of both networks to generate real-time spatial attention overlays alongside a max-weighted microvascular risk index. Evaluated on the held-out stratified validation fold (𝑛=211 scans from 𝑁=1,055 plantar skin images), the DFU branch achieves an Area Under the Receiver Operating Characteristic Curve (ROC-AUC) of 0.9985, a Macro F1-score of 0.9763, 98.06% sensitivity, and 99.07% specificity. Deployed via a decoupled FastAPI inference engine and responsive web interface, DiaGuard demonstrates a reproducible, low-latency architecture for comprehensive outpatient diabetic screening.

Zenodo (CERN European Organization for Nuclear Research)
Oldham Council (GB)
Diabetic Foot Ulcer Assessment and Management
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon