Deep Learning Driven Liver Tumor Detection and Classification

Abstract: Liver tumors represent a significant health concern globally, ranking as the fifth most prevalent type of tumor in men and the ninth in women according to recent Global Cancer Statistics for 2018. The current diagnostic methods involve utilizing imaging tests such as Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and ultrasound, often requiring invasive tissue sampling for confirmation. However, these approaches are not only costly but also time-consuming, posing challenges in timely diagnosis and treatment. This paper proposes an innovative solution leveraging deep learning techniques in image processing for liver tumor diagnosis. Specifically, we introduce a Classification Ensemble model combining ResNet, UNet, ResUNet, Xception, and DenseNet architectures. These models are adept at extracting intricate features from medical images, enhancing the accuracy of tumor classification. Additionally, we integrate state-of-the-art object detection models, including FasterRCNN, YOLOv3, and YOLOv5, to enable precise localization and delineation of liver tumors within the images. We focus on the liver tumors such as cholangiocarcinoma and metastatis. Overall, our proposed framework holds promise in revolutionizing liver tumor diagnosis, potentially leading to earlier detection, improved patient outcomes, and reduced healthcare costs.