A Radiomics-Integrated Deep-Learning Model for Identifying Radionecrosis Following Brain Metastasis Stereotactic Radiosurgery
J Zhao1*, Z Yang1, Z Hu2, E Vaios1, D Carpenter1, S Floyd1, K Lafata1, F Yin1, C Wang1, (1) Duke University, Durham, NC, (2) Duke Kunshan University, Kunshan, Jiangsu, China
Presentations
TU-I345-IePD-F5-2 (Tuesday, 7/12/2022) 3:45 PM - 4:15 PM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 5
Purpose: To develop a radiomics-integrated deep-learning(RIDL) model for identifying radionecrosis in brain metastasis patients with post-SRS MR progressions.
Methods: Utilizing the 3-month post-SRS high-resolution T1+c scan, the developed model comprises three key steps: 1) 184 radiomics features are extracted within SRS planning target volume(PTV) and 60%-isodose volume(V60%) followed with Z-score normalization; 2) a deep neural network(DNN) mimicking the encoding path of U-net is trained for radionecrosis/recurrence prediction using the 3D MR volume. Prior to the binary prediction output, latent variables in the DNN are extracted as 512 deep features; and 3) all extracted features are synthesized as an input of support vector machine (SVM) execution. Key features with prominent linear kernel weighting factor values are identified by clustering analysis, and these key features are utilized by SVM to generate the final radionecrosis/recurrence prediction result. The RIDL model was studied using a 51-patient cohort with known biopsy outcome(37 radionecrosis, 14 recurrence). During the model training, a 7:3 training/test data sample ratio was adopted, and 50 model versions were acquired with random validation sample assignments. Sensitivity, specificity, accuracy, and ROC of the model were evaluated, and these results were compared with 1)classic radiomics-based prediction (radiomics features+SVM) and 2)DNN prediction results. Wilcoxon signed-rank test with a significance level 0.05 was adopted when applicable.
Results: While radiomics-based prediction achieved acceptable accuracy (0.605±0.053) but very low sensitivity (0.130±0.125), DNN prediction achieved similar accuracy (0.567±0.006) with an improved sensitivity (0.440±0.254). In contrast, the RIDL model achieved the best prediction accuracy (0.643±0.059) and sensitivity (0.650±0.122). Additionally, ROC of the RIDL model (AUC=0.688±0.035) was superior to the other two prediction methods with statistical significance(Radiomics: 0.529±0.116; DNN: 0.637±0.055).
Conclusion: The developed RIDL model can accurately differentiate brain metastasis radionecrosis/recurrence using a single post-SRS MR scan. It holds great potential as a clinical decision-aid tool for brain metastasis outcome management.
Funding Support, Disclosures, and Conflict of Interest: This work is partially supported by institutional P30 Cancer Center Support Grant (Grant ID: NIH CA014236)
Keywords
Radiosurgery, Modeling, Image Analysis
Taxonomy
IM/TH- Image Analysis (Single Modality or Multi-Modality): Machine learning
Contact Email