First CIMILab Paper Accepted at MICCAI MIRASOL 2026
A landmark milestone for the lab — our first publication, on compute-efficient spinal lesion detection, has been accepted at the 3rd MIRASOL Workshop at MICCAI 2026 in Strasbourg, France.
Acceptance Notification
Click image to view full size
About the Paper
A Sensitivity-Gated Cascade for Compute-Efficient Spinal Lesion Detection in Resource-Constrained Settings
Prosenjit Mondol · Samrat Kumar Dey · Arpita Howlader
Most existing spinal lesion models stop at binary predictions — but clinical reality is far more dynamic. This work introduces the DERNet Framework for spinal lesion triage and localization, predicting where and which spinal lesions occur directly from a spine X-ray. The architecture couples a YOLO11-I detector with a weighted-ensemble classifier (DenseNet121, EfficientNetV2-S, ResNet50), gated by a sensitivity threshold to balance clinical recall and computational cost.
Evaluated on the VinDr-SpineXR dataset, the framework is purpose-built for resource-constrained settings — exactly the focus of the MIRASOL workshop — where datasets, infrastructure, and compute may be limited.