LFT-Transformer: FT-Transformer with Linear Dimensionality Reduction for Renal Cell Carcinoma Prediction

Early detection of renal cell carcinoma (RCC) is crucial for improving patient survival and reducing mortality. As models based on deep learning have been proven to improve the remarkable performance of disease risk prediction, yet they are mainly applied to imaging data rather than tabular data, which are frequently applied in real-world clinical practice. This study proposes a deep learning model based on the Transformer called LFT-Transformer for early RCC predictions. Aiming at a total of 2940 high-dimensional clinical and radiological features, this study uses feature embedding to convert heterogeneous features into learnable Token embeddings for effectively processing both numerical and categorical features. Furthermore, this study modifies the self-attention mechanism to compute the contextual mapping in linear time complexity, which can improve the ability to process high-dimensional structured data, making it particularly well-suited for multimodal structured inputs in this study. Experiments were conducted on a real-world medical dataset. Results highlight the model’s superiority over current state-of -the-art machine learning methods in RCC risk predictions, with significant potential to enhance RCC screening and early diagnosis.