IMPROVED LINKNET WITH TRANSFER LEARNING FOR ASSESSING MORTALITY RISK BASED ON PRETRANSPLANT DATA OF LIVER TRANSPLANTATION

Liver Transplantation (LT) is a complex and high-risk medical procedure, with post-operative mortality being a significant concern. Accurately predicting mortality risk is essential for improving outcomes, optimizing resource allocation, and guiding clinical decisions. This research introduces a novel Deep Learning (DL) model implemented based on Transfer Learning (TL), namely Modified LinkNet with Transfer Learning (mLiNet-TL). The mLiNet-TL approach is designed to predict mortality risk in liver transplant patients using pretransplant data. This approach involves several key steps, including data preprocessing, feature extraction, feature selection, and mortality risk prediction. Initially, data preprocessing is performed using a Modified Max Abs Scaling (MMAS) technique, which improves on conventional methods through an exponential transformation for greater efficiency. Essential features such as entropy-based, information gain, raw data, and statistical features are then extracted during the feature extraction step. These features are further refined through a Hybrid Weighted Feature Selection (HWFS) model, which combines the strengths of filter (by the variance threshold technique) and wrapper (by the Forward Feature Selection (FFS)) methods to identify the most relevant predictors of mortality risk. The mLiNet-TL model is implemented based on a feature-based TL approach for mortality prediction. This allows the model to effectively leverage prior knowledge for more accurate predictions. This model’s performance is compared to traditional approaches, with results indicating significant improvements in terms of accuracy, specificity, precision, etc., especially with a maximum accuracy of 96%.