Lymphocyte Segmentation Using Enhanced-YOLO in Histopathological Images of Mycosis Fungoides

Mycosis Fungoides (MF), most common cutaneous T-cell lymphoma, is diagnosed by detecting lymphocytes in epidermal region of skin in histopathological images. This study focused on detecting positive lymphocyte in the epidermal layer of the skin in patients diagnosed with MF. To this end, a set of pathology slides was first collected and annotated under the supervision of a pathologist. The YOLO11 model was initially evaluated on a public dataset and demonstrated superior performance compared to other existing models. Based on these results, YOLO11 was then applied to train on our collected dataset to detect lymphocytes within the epidermal region. While YOLO11 demonstrated strong detection capabilities, further refinement was required to enhance its precision and reduce false positives and false negatives, particularly in complex or overlapping regions. To address this, an Attention U-Net was employed as a restoration network, specifically designed to improve the YOLO11 predictions. This combination of YOLO11 for initial detection and Attention U-Net for post-processing allowed for improved localization accuracy, better detection of lymphocytes, and more reliable classification across different tissue regions.