Steatosis Assessment in Liver Transplants Using Transformers for MASH Classification

An accurate classification of Metabolic dysfunction-associated steatohepatitis is crucial to assess donor organ quality and post-transplant outcomes in liver transplants. However, it currently struggles with challenges such as shortage of expert pathologists, high inter-observer variability and a time-consuming diagnosis process. Thus, to counter these challenges, we present a fully automated algorithm to classify hepatic steatosis in liver transplants by using a four-step deep learning based approach. The first step utilizes the SegFormer B4 model for semantic segmentation to obtain histological tissue detection. Next, we perform initial steatosis segmentation with the help of Gabor kernels, followed by an adaptive color deconvolution to separate hematoxylin and eosin channels along with the Macenko method for extended steatosis detection. Finally, the cleaning steps involve the use of SegFormer B3 model to perform semantic segmentation to reject false positives in macrosteatosis, while logistic regression is used for microsteatosis utilizing factors such as size, roundness, circularity, and solidity. The algorithm is developed and tested upon 385 H&E stained liver images obtained from 77 donors. Our algorithm obtains a 99.44% train and 99.41% test accuracy for the tissue detection, and a 97.54% train and 97.11% test accuracy for steatosis, thus outperforming the existing techniques.