Automatic segmentation of liver tumors from computed tomographic images using hybrid deep learning model

The liver isa vital organ which is responsible for numerous critical functions and it is prone to various diseases, with liver cancer being one of the deadliest. Liver cancer is the third most common cause of cancer-related fatalities worldwide, taking a considerable number of lives. Effective therapy depends on early diagnosis, especially for primary liver cancers like hepatocellular carcinoma (HCC). Early detection depends on the ability to segment the liver and its tumor from abdominal Computed tomography (CT)images, but this is a very difficult process because of the liver’s complicated structure, different sizes, irregular boundaries and similar intensity values to those of other organs. In this work we created a strong deep learning model to tackle these issues called RV-UNet which automatically selects and processes features using convolutional layers while preserving the spatial information of the extracted features. The Liver Tumor Segmentation (LiTS) dataset was used to test the RV-UNet model, which concentrated on primary HCC (Hepatocellular Carcinoma) images that had been resized to 128 × 128. The proposed model achieved a Dice Similarity Coefficient of 88.96% , a Jaccard Index of 80.59% and a Volumetric Overlap Error of 20.38% for the segmentation of liver tumors on the LiTS dataset. These findings show that the model outperforms the most advanced techniques for liver tumor segmentation in abdominal CT scans, especially when the image size is 128 × 128.