Abstract
Automated analysis of spinal radiographs is essential for early diagnostic triage but remains challenging due to the visual subtlety and small scale of many spinal lesions. We propose a unified cascaded deep learning framework designed to improve screening sensitivity while maintaining precise lesion localization. The diagnostic workflow is structurally decoupled into two stages. First, DERNet, a heterogeneous ensemble of EfficientNetV2-S, DenseNet121, and ResNet50, performs high-sensitivity binary triage to filter normal radiographs. Abnormal cases are subsequently routed to a customized YOLO11-L detector for fine-grained lesion localization. An experimental evaluation of the VinDr-SpineXR benchmark demonstrates strong performance, achieving an AUROC of 91.03% for image-level classification and a Mean Average Precision (mAP@0.5) of 40.10% for lesion-level detection. To enhance clinical interpretability, we integrate explainability techniques, including saliency map, Grad-CAM, and LIME, ensuring that predictions align with anatomically relevant structures. Our proposed framework, designed for low-resource settings, transforms standard 2D X-rays into a highly efficient smart triage system, ensuring that critical patients receive strict priority for scarce MRI resources. Additional resources, including experimental analyses, error analysis, deployment interface, and source codes, are available at https://cimilab.github.io/DERNet/.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_074.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=91seetK1GO
BibTex
@InProceedings{MonPro_ASensitivityGated_MICCAISAT2026,
author = { Mondol, Prosenjit AND Dey, Samrat Kumar AND Howlader, Arpita},
title = { { A Sensitivity-Gated Cascade for Compute-Efficient Spinal Lesion Detection in Resource-Constrained Settings } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17264},
month = {pending},
page = {pending}
}
