MULTICLASS U-NET FOR LIVER AND TUMOR SEGMENTATION IN ABDOMEN CT SCANS

COMPUTER ENGINEERING

Liver tumor segmentation in CT(Computed Tomography) scan images plays a critical role in diagnosis, treatment planning and monitoring of liver diseases. In recent years, deep learning techniques, particularly the U-Net architecture, have shown remarkable success in medical image segmentation tasks. However, segmenting both liver and tumor regions accurately in CT scans presents unique challenges due to variations in shape,size and intensity levels of lesions.This paper presents a novel multiclass U-Net architecture designed specifically for liver tumor segmentation in CT scan images. The proposed model integrates both liver and tumor classes into a unified segmentation framework, enabling simultaneous extraction of relevant anatomical structures and pathological regions. The U-Net architecture is well-suited for this task, as it effectively captures spatial dependencies and hierarchical features within the input images.Key components of the proposed multiclass U-Net include an encoder-decoder structure with skip connections for feature fusion, convolutional layers with batch normalization and non-linear activations, and a final softmax layer for pixel-wise classification into liver and tumor classes. The model is trained using a large dataset of annotated CT scans, leveraging techniques such as data augmentation and transfer learning to improve generalization performance.The proposed multiclass U-Net achieves state-of-the-art performance in liver tumor segmentation tasks, accurately delineating both liver and tumor regions across different patient cohorts.The experiments demonstrated that this method can accurately segment liver tumors.We achieved True value Accuracy of up to 98.4%.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.