Liver Tumor Detection Using CLAHE Histogram Equalization and U-Net Model from 3D CT Scan Images

Automating liver tumor detection is crucial for early diagnosis and effective treatment planning. This paper leverages deep learning techniques, specifically the U-Net architecture and image processing techniques like CLAHE, to detect liver tumors. It employs the LiTS17 dataset, comprising 3D CT scan images processed with various techniques for model training. The trained U-Net model exhibits exceptional performance in liver segmentation and tumor detection, accurately localizing tumors. Initially, the liver is segmented, and then the same model is used for detection. Evaluation results demonstrate high accuracy, with liver detection achieving an IoU Score of 91.0% and a Dice Score of 86.4%. Additionally, liver tumor detection achieves an IoU Score of 82.7% and a Dice Score of 92.4%. Comparative analysis with five models reveals the proposed model’s superior accuracy in detecting liver tumors. These findings highlight the potential of deep learning for automated liver tumor identification, facilitating early intervention and treatment planning.