Differentiating multiple myeloma and bone metastasis on spine MRI using radiomics-based machine learning and deep learning models

Motivation: Spinal bone metastasis and multiple myeloma share similarities in MRI patterns and clinical symptoms, making their differentiation challenging. Goal(s): To evaluate the feasibility of radiomics-based machine learning and deep learning models in distinguishing between multiple myeloma and metastasis on T2-weighted MRI. Approach: Radiomics-based random forest and two deep learning architectures that utilized the recurrent residual convolutional neural network and the spatial attention mechanism were developed for the prediction models. Results: The proposed models demonstrated the potential of differentiating between two challenging conditions, with an accuracy of 75.0% and an AUC of 0.804. Impact: The results from this study suggest that the proposed approach may provide a valuable assisting tool in distinguishing two challenging conditions, MM and metastasis on spine MRI.