International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Object Detection on Detecting Skin Lesion using Dab-DETR

Authors :

Sheida Shadman¹ Amirreza Rouhbakhshmeghrazi² Shayan Nalbandian³ Bo Li⁴ Shaghayegh Shadman⁵ Malik Muhammad Owais Siddique⁶

1- School of Software Engineering Northwestern, Polytechnical University, Xi’an, China 2- School of Electronics and Information Northwestern Polytechnical University Xi’an, China 3- School of Software Engineering Northwestern Polytechnical University Xi’an, China 4- School of Electronics and Information Northwestern Polytechnical University Xi’an, China 5- Arts Faculty Alzhara University Tehran, Iran 6- School of Software Engineering Northwestern Polytechnical University Xi’an, China

Keywords :

Medical imaging،Vision Transformer،Health care AI،Dynamic Anchor Boxes،Small Object Detection

Abstract :

—Melanoma is one of the deadliest forms of skin cancer, where timely and accurate detection is critical. However, traditional object detection models struggle with small, irregular lesions and slow convergence—common challenges in dermo scopic imaging. This study aims to enhance melanoma detection by using DAB-DETR, a DETR-based model that incorporates dynamic anchor boxes to improve spatial awareness and training efficiency. We trained and evaluated the model on a curated dermoscopic dataset of 650 annotated images (four melanoma related classes), using PyTorch and standard evaluation metrics. DAB-DETR’s performance was compared to standard DETR, Deformable DETR, RetinaNet, and Faster R-CNN. The results show that DAB-DETR achieves an AP of 54.2%, outperforming Deformable DETR (51.8%) and showing improvements at AP50 (61.3%) and AP75 (55.1%). Recall increased from 73.7% to 75.0%, demonstrating better sensitivity to hard-to-detect cases. The findings highlight the model’s effectiveness in improving lesion localization and convergence, addressing key limitations of existing DETR-based methods. This makes DAB-DETR a more practical and accurate tool for automated melanoma detection in clinical settings.