Differentiating Multiple Myeloma and Osteolytic Bone Metastasis on Spinal CT Scan: A Comprehensive Study Using Convolutional Neural Network

Background: Accurate differentiation of spinal multiple myeloma (MM) and osteolytic metastatic bone tumor (OMBT) can be challenging. Usually, imaging methods, laboratory tests, and biopsy are performed for the correct diagnosis. In this study, we aimed to differentiate CT images from patients with MM and OMBT using CNN models. Materials and Methods: 3707 CT images of 91 patients (1886 OMBT images and 1821 MM images; 46 males and 45 females; mean age: 61.2 years) obtained between January 2015 and January 2023 were reviewed. 2667 images were randomly selected for the training set, 740 for the validation set, and 300 for the test set. A transfer learning approach was used based on DenseNet121, DenseNet169, EfficientNetB0, MobileNet, MobileNetV2, VGG16, and Xception CNN architectures. The performance of the models was evaluated. Results: When the sensitivity, specificity, positive predictive value, negative predictive value, accuracy, F1 score, and kappa measurements of the models in the MM and OMBT differentiation are evaluated, the most successful ones are MobileNetV2, MobileNet, and VGG16, with accuracy of 88%, 86.33%, and 86%, respectively. Conclusions: Our study showed that CNN-based artificial intelligence models can differentiate MM and OMBT on CT images.