Automated Liver Segmentation in MR T1 In-Phase Images Transfer Learning Technique

Liver conditions are on the rise due to a variety of factors such as alcohol consumption, diabetes, and obesity, which require expert diagnosis and treatment. The segmentation of medical images is vital for achieving accurate diagnoses as it effectively distinguishes between different image types across various modalities. This paper focuses on liver segmentation of 2D MR T1 in-phase abdominal images using the U-Net architecture, which employs CNN layers and an encoder-decoder model to extract features and achieve precise localization. The U-Net architecture leverages both up-sampling and down-sampling operations to achieve highly accurate liver segmentation. The dataset employed in this study is sourced from the CHAOS grand challenge, comprising 2D MR T1 in-phase abdominal images acquired from healthy subjects. In our experiments, the U-Net model achieved remarkable results, with a Dice score of 96% for liver segmentation in MR T1 in-phase images. This demonstrates the effectiveness of U-Net in accurately delineating liver regions, which is vital for the precise diagnosis and treatment of liver diseases.