Enhancing Liver Tumor Segmentation with Novel UNet++ Hybrid Models: ResNeXt, ResNet and InceptionV3

Our work presented novel hybrid models for liver tumour segmentation that use UNet++ with other backbone designs, such as ResNet, ResNeXt, and Inception networks. These new hybrid methods were thoroughly assessed to determine their effectiveness in this particular type of specialized medical imaging. Out of all of these hybrid models, UNet++ in conjunction with the ResNeXt backbone notably produced consistently excellent results, achieving a high IoU score of 0.8683, and a small loss of 0.1182, a remarkable F1 score of 0.8535. This work represents a noteworthy advancement in the field of medical image analysis by emphasizing the ability of novel model combinations to tackle complex problems related to liver tumour segmentation.