Leveraging deep learning and explainable AI for effective liver tumor classification from CT scan images.

Introduction: Cancer remains one of the leading causes of mortality worldwide, with 19.3 million new cases and 10 million deaths reported in 2020. According to the World Cancer Research Fund International (WCRF), liver cancer ranks as the fifth most prevalent cancer in men and the ninth in women. Despite available interventions, liver cancer is often diagnosed at advanced stages due to its subtle progression and the complexity of distinguishing hepatic malignancies from surrounding tissues in CT scans. Conventional diagnostic practices, such as biopsy, are invasive, time-consuming, and mentally exhausting for patients, while manual interpretation of CT images is labor-intensive and requires expert radiologists. These challenges highlight the urgent need for automated, accurate, and explainable diagnostic tools. Methods: In this work, we propose a comprehensive deep learning framework for non-invasive liver tumor classification with integrated explainability. We evaluated and fine-tuned several state-of-the-art supervised models, including ResNet50-v2, EfficientNetV2, Inception-v3, and Vision Transformer ViT-16, combined with tailored pre-processing and augmentation strategies. Results and discussion: The EfficientNetV2 model achieved 96.97% accuracy, demonstrating competitive performance with existing literature. Beyond high accuracy, the framework integrates explainable AI methods to enhance interpretability and clinical trust, bridging a key gap in current AI-driven liver cancer research.