Evaluating a Machine Learning Tool for the Classification of Pathological Uptake in Whole-Body PSMA-PET-CT Scans

The importance of machine learning (ML) in the clinical environment increases constantly. Differentiation pathological from physiological tracer-uptake positron emission tomography/computed tomography (PET/CT) images is considered time-consuming and attention intensive, hence crucial for diagnosis treatment planning. This study aimed at comparing validating supervised ML algorithms to classify uptake prostate cancer (PC) patients based on prostate-specific membrane antigen (PSMA)-PET/CT. Retrospective analysis 68Ga-PSMA-PET/CTs 72 PC resulted a total 77 radiomics features 2452 manually delineated hotspots training labeled (1629) or (823) as ground truth (GT). As held-out test dataset, 331 (path.:128, phys.: 203) were 15 other patients. Three classifiers trained ranked assess classification performance. result, high overall average performance (area under curve (AUC) 0.98) was achieved, especially detect (0.97 mean sensitivity). However, there still room improvement (0.82 specificity), glands. algorithm applied lesions predicts hotspot labels with accuracy unseen data may be an important tool assist diagnosis.