Adapting YOLOv8 for Kidney Tumor Segmentation in Computed Tomography

Kidney tumors pose a significant health challenge, ranking among the top ten cancers affecting both men and women globally. Early diagnosis of kidney neoplasms improves treatment outcome and survival rates by preventing advanced stages of cancer along with lower occurrence of metastasis. The advent of medical imaging has been a cornerstone in the fight against such cancers, enabling the development of algorithms that assist healthcare professionals in diagnosis and treatment planning. Machine learning (ML) models, in particular, have shown promise in enhancing the accuracy and efficiency of medical diagnoses. The paper leverages a general-purpose pretrained ML model, specifically adapted, and fine-tuned for the segmentation of kidney tumors using the KiTS23 dataset. The extraction of 2D images is performed from CT scans, which are more suitable for this training approach. The pretrained YOLOv8n model is employed which is appropriate for real-time segmentation. The model is evaluated using precision, recall, mAP50, mAP50-95, and Sørensen-Dice coefficient achieving
$0.94,0.89$
,
$0.96,0.72$
, and 0.79, respectively. These results show that the application of YOLOv8 model for the segmentation of kidney tumors has proven to be a judicious choice. This approach has the potential to significantly benefit both patients and physicians by improving diagnostic accuracy and treatment outcomes. These findings lay a solid foundation for future studies aimed at integrating advanced ML models into healthcare settings, ultimately enhancing patient care.