Enhanced identification of pheochromocytoma and paragangliomas' genetic clusters from CT

Pheochromocytoma and Paraganglioma (PPGL) tumors are uncommon neuroendocrine tumors that exhibit substantial genetic and phenotypic variability, making their classification challenging. We describe a new method that performs supervised contrastive learning leveraging DinoV2 feature representations to categorize genetic clusters of PPGL tumors into four distinct subtypes: SDHx, VHL-EPAS1, kinase signaling, and sporadic. The proposed supervised contrastive learning framework teaches a shallow linear network using feature representations taken from a DinoV2 model that has already been pre-trained with ImageNet-21K. This training process optimizes both contrastive and classification losses together, taking into account the genetic groups and four anatomical placements of PPGLs: head and neck, chest, adrenal, and extra-adrenal abdomen and pelvis. The proposal was tested on a dataset consisting of 650 contrast-enhanced computed tomography (CE-CT) axial images of 287 PPGL patients. The results showed a balanced accuracy of 0.582±0.079, an AUC score of 0.805±0.055, and an F1 score of 0.551±0.063 using five-fold cross-validation. Our methodology demonstrated superior performance compared to other methods by balancing the performance across all four genetic types, achieving a 2% higher balanced accuracy and a 7–14% higher F1 score, while using the entire axial slice as a single sample in contrast to using the extracted tumor patches. Moreover, the sporadic cluster achieved the highest balanced accuracy (0.652±0.037), while the SDHx cluster had the highest F1 score (0.738±0.0565). Experimental studies have verified that incorporating predictions that are aware of specific regions during training, as well as utilizing supervised contrastive learning, are both essential for attaining these outcomes.