CBAM and ECA Attention Mechanisms in YOLOv10n for Diabetic Foot Ulcer Detection

Diabetic foot ulcers (DFU) are one of the most common and costly complications of diabetes, which, if not diagnosed and treated promptly, can lead to severe infections and ultimately amputation. Early and accurate localization of ulcers in clinical images is therefore critical for any automated monitoring system. In this paper, we enhance the lightweight YOLOv10n model for DFU detection by comparatively integrating two efficient attention modules—CBAM and ECA—separately into the neck architecture. Both modules are inserted at three specific points using a residual gated design with learnable scaling factors to ensure stable training without disrupting the pretrained backbone. We evaluate three configurations (baseline, +CBAM, and +ECA) on the public subset of the DFUC2022 dataset (2000 images) under identical training protocols. Experimental results show that adding CBAM increases the accuracy of mAP50 from 77.42% to 81.48% (an improvement of 4.06 percentage points) and the Recall criterion by about 5.5 percentage points, while ECA records the largest improvement in Recall (6.4 percentage points) and mAP50-95 but is accompanied by a relative decrease in Precision. Both modules provide a noticeable improvement in the detection of small wounds and irregular borders with less than 0.17% increase in model volume. These results indicate that lightweight attention mechanisms can improve the performance of nano YOLO models in medical imaging applications without imposing significant computational cost.