Vision Mamba for Liver Tumor Diagnosis in Multi-phase Magnetic Resonance Imaging

Motivation: Early and accurate diagnosis of liver tumors is crucial for effective treatment and prognosis. Goal(s): We aim to develop a deep learning model for accurate diagnosis of primary liver tumors in multi-phase MRI. Approach: We propose a Vison-Mamba based deep learning network for liver tumor classification, leveraging the strengths of State Space Models (SSM) combined with convolutional operations to capture long-distance dependencies. Results: We conducted the evaluation with 104 patients. Experiment results demonstrate our proposed model achieve the highest accuracy in diagnosing seven types of primary tumors among the methods compared in this study. Impact: Our proposed deep learning method can diagnose most primary tumors with high accuracy. It has the potential to benefit treatment planning and improve patient outcomes.