Enhanced Liver Image Segmentation Using Custom Metric and Unet with FastAI

Liver segmentation from medical images is a crucial task in computer-aided diagnosis for detecting liver abnormalities and supporting clinical decision-making. However, accurate segmentation remains challenging due to variations in liver shape, intensity, and the presence of surrounding organs. In this paper, an enhanced liver image segmentation approach is proposed using the U-Net deep learning architecture implemented with the FastAI framework. The proposed method utilizes efficient preprocessing techniques and data augmentation to improve model generalization. In addition, custom evaluation metrics, including the Dice Similarity Coefficient and Intersection over Union (IoU), are incorporated to better measure segmentation performance and optimize the training process. The model is trained and validated on liver medical image datasets to accurately identify and segment liver regions. Experimental results demonstrate that the proposed approach achieves improved segmentation accuracy and robustness compared to conventional segmentation techniques. The integration of FastAI with U-Net simplifies model development while maintaining high performance, making the proposed method suitable for real- time clinical support and automated medical image analysis.