Predicting Liver Tumor: Leveraging Image Processing with DenseNet121

• Published in: 2024 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI) pp. 1 - 8
• Main Authors: B, Sandhiya, Selvan, R. Anbu, Gowtham, R.N.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 09.05.2024
• Subjects: Accuracy; Computed tomography; Confusion matrix; Convolutional neural network; Deep learning; Densenet121; Image processing; Image segmentation; Lesions; Liver; Predictive models; Semantic segmentation
• ISBN: 9798350389432
• DOI: 10.1109/ACCAI61061.2024.10601912

Early detection and accurate prediction of liver disease play a crucial role in improving patient outcomes and reducing the burden on healthcare systems. Segmenting the liver and its tumors using computed tomography (CT) images is an essential undertaking for the diagnosis and treatment of liver illnesses. Because of the uneven distribution, hazy boundaries, varied densities, forms, and sizes of lesions, segmenting the liver and associated tumor is a challenging task. Up until this point, our primary focus has been on developing deep learning algorithms that can separate the liver and its tumor from CT scan pictures of the abdomen, saving time and effort when diagnosing liver illnesses. A deep learning-based automatic segmentation method is presented that uses the improved densenet121 model to segment the liver and its tumor. In this model image processing is used for the accurate automated segmentation of tumors, the proposed method demonstrates the ability to accurately segment the liver as well, as indicated by the confusion matrix obtained when comparing to the previous work on liver and tumor segmentation. The Densenet121 architecture serves as the foundation for the algorithm employed here, we introduced an autonomous technique to segment the liver from CT scans and lesions from the segmented liver region, based on semantic segmentation convolutional neural networks (CNNs). For liver and tumor segmentations, respectively, the suggested system achieves an accuracy of 95.31% to 95.39%.