361P - Radiomic model predicting radiological response after thoracic stereotactic body radiotherapy regardless of tumor histology and staging

Background

Thoracic stereotactic body radiotherapy (SBRT) is widely applied in both early and metastatic disease. Pathological CR rate after SBRT was quoted around 60%. Thus, it is important to predict responder and non-responder to SBRT. With advent of radiomics, textual features of tumor can be extracted from imaging. We propose a model to predict radiological response after SBRT based on tumor radiomics features regardless of histology and staging.

Methods

Patients receiving thoracic SBRT using active breathing control (ABC) were retrospectively recruited regardless of tumor histology/primary and staging. All patients received 50-54 Gy in 3-4 fractions equivalent to BED >100Gy. All patients had regular contrast CT Thorax per protocol and PET/CT if indicated. Tumor response was assessed by an independent senior radiologist based on RECIST criteria. Responders are defined as complete response (CR) or partial response (PR). Non-responders were defined as those with stable or progressive disease. Gross tumor volumes (GTV) were contoured on the initial planning CT. 110 radiomics features including voxel intensities, textual and gray level features were extracted using pyradiomics module. The features were then analyzed using in-house software. A model using support vector machine (SVM) was trained to predict response based solely on the extracted radiomics features. 10-fold cross validation was used to avoid overfitting. ROC curves were constructed to evaluate model performance.

Results

68 patients were recruited from 2008 to 2018. 54 patients had lung primaries while 14 patients had thoracic oligo-metastases. Secondaries include colorectal, head and neck squamous cell carcinoma and hepatocellular carcinoma. 85 tumors were analyzed, of which 31 tumors had CR and 11 tumors had PR. The radiomic model developed had an accuracy of 74.8%. The AUC for CR, PR and non-responder prediction was 0.865 (95% CI: 0.794 – 0.921), 0.946 (95% CI: 0.873 – 0.978) and 0.857 (95% CI: 0.789 – 0.915) respectively. Under the threshold, the sensitivity was 89% while the specificity was 68% for detecting non-responders.

Conclusions

Radiomic is a promising technique that can predict tumor response with good accuracy.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Department of Clinical Oncology, Queen Mary Hospital.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.