Deep Learning-Based Diagnostic System for Liver Tumor Classification Using Dynamic Contrast-Enhanced MRI Images Combined with Clinical Data

A deep learning based diagnostic system is proposed in this research for classifying liver tumors, using dynamic contrast-enhanced MRI images with clinical data. This research integrates Inception V3, ResNet 50 and VGG Input which are optimize deep learning models together RNN to process the images input as well clinical data or actual report. Extracted 3,200 records were provided in a dataset with splited as training set (70%) and test set (30%). All MRI images and clinical information underwent a comprehensive preprocessing step to prepare ideal input quality for model training. Every model was assessed against the level of accuracy, precision and recall from F1 score. Inception V3 model with RNN was the most accurate among all models correctly predicting 97.8% of patients, followed by ResNet 50 (94.5%), and lastly VGG 16 (91.2%). The fact that the classifier Inception V3 with RNN showed best performance in classifying liver tumors demonstrates its superiority to AI system for clinical diagnostics. Our novel integrated strategy paves the ways for deep learning models have great potential to improve diagnostic performance when both of imaging and clinical data are used, which can provide a more powerful method in liver tumor classification. Overall, the findings further expand on increasing efforts of utilizing AI in diagnostics and shed light to improve early diagnosis on liver tumor detection.