A random forest radiomics model achieved satisfactory diagnostic performance, and models based on joint CET1 and T2-weighted imaging demonstrated better performance than single-sequence models in differentiating spinal multiple myeloma from metastases.
Spinal metastases and spinal multiple myeloma have similar radiographic characteristics, making accurate differentiation a necessity for precision therapy.
This retrospective study established and assessed different radiomics models, including logistic regression (LR), AdaBoost, support vector machines (SVM), random forest (RF), and multiple kernel learning-based SVM (MKL-SVM), to achieve accurate differentiation between the two conditions. The findings are reported in the Journal of Bone Oncology.
Baseline Characteristics
The two-center retrospective study included 263 patients, with 127 diagnosed with multiple myeloma and 136 diagnosed with metastases. The tumor origins in spinal metastasis patients included the lung in 72 patients, breast in 14 patients, liver in 22 patients, kidney in 24 patients, and gastrointestinal tract in 4 patients.
Classification Performance of Radiomics Models Based on T2 Alone and CET1 Alone
Based on contrast-enhanced T1-weighted imaging (CET1) alone and T2-weighted imaging (T2WI) alone, the random forest (RF) radiomics model outperformed other models in internal training and external validation sets. The T2WI models were found to be superior to models based on CET1.
Classification Performance of Radiomics Models Based on Both T2 and CET1
The classification performance of radiomics models based on T2WI and CET1 models was comparatively better than the performance of models based on T2WI and CET1 alone. Accuracy and area under the ROC curve were 0.862 and 0.870, respectively, for the MKL-SVM model; 0.853 and 0.865, respectively, for the RF model; 0.823 and 0.833, respectively, for the SVM model; 0.791 and 0.793 for the AdaBoost model; and 0.752 and 0.788 for the LR model.
Source:
Cao, J., Li, Q., Zhang, H., Wu, Y., Wang, X., Ding, S., Chen, S., Xu, S., Duan, G., Qiu, D., Sun, J. F., Shi, J., & Liu, S. (2024). Radiomics Model based on MRI to Differentiate Spinal Multiple Myeloma from Metastases: A Two-center Study. Journal of Bone Oncology, 45, 100599. https://doi.org/10.1016/j.jbo.2024.100599
