Enhanced ResNet Model for Classification of Liver Tumor in CT Scans

Liver Tumor is one of the common and highly occurring health issue worldwide with an extremely high mortality rate. Early diagnosis is the reason behind increasing the survival rate of patients through treatment at the proper time. Deep Learning (DL)-based techniques have shown extremely promising results in recent years for medical image analysis, particularly tumor classification. But traditional techniques are marred with the issues of feature extraction, generalization, and classification accuracy. Here, we introduce an optimized ResNet50 model for binary liver tumor classification from CT scans. Fine-tuning the model makes it learn more discriminative features, hence improving the classification accuracy. Our steps include preprocessing CT scans, hyperparameter tuning, and training the model on a well-processed dataset. For the comparison of the performance of our strategy, we perform a stringent performance evaluation based on different evaluation metrics such as accuracy, precision, recall, and F1-score. The outcome reveals that our fine-tuned ResNet50 achieves greater performance compared to traditional feature extraction techniques with better classification accuracy. The acceptable model performance validates its prospect as a radiologist-assistance system for early liver tumor detection to alleviate diagnostic errors and promote clinical decision-making. Our research contributes to AI-aided diagnosis in medicine and opens the door for further studies on automated liver tumor classification.