ResNet50-SCPA-net for liver cirrhosis stage classification using MRI images

Liver cirrhosis is a chronic liver disease that develops in different stages and may cause serious complications if not diagnosed early. However, identifying the stage of liver cirrhosis from MRI images is difficult because the appearance of the liver changes as the disease progresses.

In this study, a modified ResNet50-based model, called ResNet50-SCPA-Net, was developed for liver cirrhosis stage classification using liver MRI images. The proposed model combined SE channel attention and CBAM spatial attention in a parallel structure to refine the extracted features. A learnable fusion parameter was used to combine the attention features during training. To evaluate the proposed framework, a comparative analysis, an ablation study, and 5-fold cross-validation were performed. Grad-CAM visualization was applied to identify the image regions involved in the prediction process.

The experimental results showed that the proposed model achieved an accuracy of 80.82%, a balanced accuracy of 81.93%, an F1-score of 82.55%, and an AUC of 93.05%. The 5-fold cross-validation produced a mean accuracy of 71.54%, balanced accuracy of 73.57%, and F1-score of 73.13%.

Overall, the findings suggest that the proposed framework can be used to automate liver cirrhosis stage classification from MRI images. Future studies should focus on larger datasets and additional imaging modalities for further evaluation.