MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Machine Learning Based On CT Radiomic Features Can Predict Residual Tumor From Radiation Changes in Head and Neck Cancer Patients Treated with Definitive Chemoradiotherapy

E Florez*, T Thomas, C M. Howard, H Khosravi, J Storrs, S Lirette, A Fatemi, University of Mississippi Medical Center, Jackson, MS

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose:
To predict residual tumor from radiation changes and inflammation through machine learning (ML) models based on radiomic features (RF) extracted from CT images.

Methods:
HIPPA-compliant, IRB-approved retrospective analysis of patients with squamous cell carcinoma of the head and neck (HNSCC) treated with chemoradiotherapy (chemoRT) at UMMC. Thirty-six patients with residual disease on CT scan performed in 2 months interval- either in primary site, nodal station or both were enrolled. Figure-1 shows the flowchart of HNSCC patients treated with chemoRT. All gross tumor volumes (GTVs) were transferred from the treatment planning CT scan to MIM® Software. Then, a radiologist contoured the tumors in 2 months follow-up CT2 and 3 months follow-up PET/CT scan using MIM’s tools. Next, the segmented GTVs from CT images were exported to MatLab® where RFs were extracted through different approaches: (a) from the region of interest (ROI) which best represented the GTV (RF2=280), (b) from the volume of interest (VOI) of the GTV (RF3=455). Finally, ML models such as support vector machine (SVM), neural network (NN) and, random forest (RaF) were used to predict changes and progress in HNSCC cancer patients treated with chemoRT.

Results:
ML models used all RF extracted from ROIs and VOIs (Figure-2). For 2D scheme, RF extracted from CT2 had the predictive ability to anticipating residual lesion in PET/CT exam (AUC=0.702). For 3D scheme, predicting positive path for residual tumor from CT2 and PET/CT had good and moderate ability (AUC=0.720 and 0.678, respectively). ML models used 10-fold cross-validation for tuning parameter optimization. Variable importance metric >80 was used to select which features contribute most to the predictive ability of each model (Table-1).

Conclusion:
ML models based on RF of CT images were able to predict the residual tumor from radiation changes in a small group of HNSCC cancer patients treated with chemoRT.

Keywords

CT, Texture Analysis, Feature Selection

Taxonomy

IM- CT: Radiomics

Contact Email