Multimodal Machine Learning-Based Cancer Progression Prediction from Plain Radiographs and Clinical Data

Abstract

This research investigates the use of a multimodal machine learning model to predict cancer progression by integrating radiographs and clinical data. The study addresses the limitations of unimodal approaches, which often overlook the synergistic potential of combining diverse data types. By leveraging deep learning techniques for image analysis and interpretable models for clinical data, the proposed framework enhances prediction accuracy and model interpretability. The multimodal model achieved a high training accuracy of 98.01% and a testing accuracy of 94%, significantly outperforming unimodal models like SVM and CNN. Precision (94.2%) and recall (94%) highlighted the model's ability to accurately identify true positive cases, while the AUC-ROC of 98% underscored its robust diagnostic capability. Comprehensive evaluation demonstrated that the multimodal model effectively integrates complementary data, improving predictive performance and supporting personalised treatment planning. The research contributes to advancing cancer diagnosis and prognosis, offering a promising tool for clinical decision-making.