Real-Time Skin Lesion Detection: An Efficient Deep Learning Approach

A crucial objective within dermatology includes the identification of skin lesions, which can help with highlighting diseases amongst the likes of melanoma early into the heart of the diagnosis process. With time, professionals have depended on manually finding skin lesions on the body – a process which is inefficient for time-management and is openly vulnerable to incorrect results. This paper suggests a deep learning-dependent methodology utilizing You Only Look Once version 8 (DLBA- YOLOv8) in the case of skin lesion detection with a quick and error-less methodology. Using the ISIC dataset, the model was effectively trained with a collection of data of dermoscopic imagery of skin lesions. With state-of-the-art integration and performance in lesion localization and classification, results from experiments have demonstrated YOLOv8’s capability to achieve a mean Average Precision (mAP) of 98.8%, comparatively overcoming the efficiency of YOLOv7 (89.3%) and Faster R-CNN (87.1%). In addition to that, when placing the results of YOLOv5 and Mask R-CNN in terms of Intersection over Union (IoU) scores side by side with YOLOv8, (85.2% and 84.6% respectively), YOLOv8 shows a great outperformance of both with a score of 88.7%. With the utilization of the Ultralytics framework and an assessment through metrics along the likes of mAP and IoU, the system shows a successful implementation. Handing the industry, a greatly developed perk in precision and overall efficiency over current existing methodologies, this paper underscores the potential capabilities of YOLOv8 in real clinical applications for skin lesion detection through an automated process.