The accurate liver segmentation and fibrosis staging are of great importance in the diagnosis of liver disease and the subsequent treatment planning. To quantify and analyze liver fibrosis, the CARE 2024 challenge proposes a multi-task track, called LiQA, which aims to automatically segment the liver from single-phase MRI scan (LiSeg task) and predict liver fibrosis staging based on multi-phase and multi-center MRI scans (LiFS task). Both the LiSeg and LiFS tasks are subject to domain shifts. Furthermore, the LiFS task may encounter challenges, including the random absence of sequences for certain patients and misalignment among multi-phase MRIs. To address these challenges, we put forward a learning-based method for liver segmentation that utilizes both extended labeled and unlabeled data. In order to address the challenges posed by the LiFS task, we have developed a three-step approach. This approach includes the cropping of liver VOIs, the identification of liver fibrosis staging using CAM regularization and the integration of multisequence results. The efficacy of our solution was evaluated using the LiQA validation and testing set, and our team achieved the best performance award in the CARE-LiQA challenge.