Real-time Prostate Cancer Detection via YOLO-Tiny Variants

Prostate cancer (PCa) stands as the most prevalent cancer diagnosis and ranks as the second leading contributor to cancer-related mortality among men worldwide. This research introduces a swift and accurate detection technique based on YOLO tiny models for identifying aberrant prostate cancer cells. With the use of this approach, aberrant cells can be promptly and precisely located and identified, enabling timely intervention and improved patient outcomes. Moreover, in pursuit of realizing an integrated deep prostate cancer detection system (PCDS), the proposed framework leverages the Nvidia Jetson Nano developer kit to evaluate the real-time efficacy of YOLO tiny detectors. The investigation draws upon a dataset comprising 3585 instances for both training and testing phases. Experimental results demonstrate that YOLO v8s achieves a mean average precision (mAP) of 99%, Precision of 99.5%, Recall of 99% and F1-score of 99.2% which indicates the effectiveness of this approach in achieving real-time prostate cancer detection thus holding great promise for enhancing medical diagnostics. When considering single-image predictions, the inferred time on the edge computing device showcases slight variations, ranging from 128ms with YOLO v5s to 32ms using YOLONASs.