Radiomic features associated with recurrence in a case:control cohort of surgically treated HPV+ OPSCC

Purpose/Objective(s) Despite the more favorable prognosis of HPV-associated OPSCC, a subset of these patients are not cured by initial therapy. At presentation, conventional staging and smoking history have poor ability to predict recurrence and survival of the HPV+ OPSCCs undergoing transoral robotic surgery (TORS) followed by pathology-guided adjuvant therapy. In order to enhance prediction of oncologic outcomes under this treatment paradigm, we evaluated a machine learning approach to identify radiomic characteristics that distinguish the patients who subsequently recurred. Materials/Methods From a single-institution cohort of patients undergoing TORS for HPV+ OPSCC from 2007 through 2017, we identified 42 patients who experienced disease recurrence within the radiation field (n=7), at distant sites (n=29), or both (n=6). These patients were matched on several characteristics (8th edition AJCC pathologic T and N stage, smoking history, adjuvant radiation, and adjuvant chemotherapy) to 42 patients from the same cohort who did not recur with minimum follow up beyond the latest recurrence. Using baseline diagnostic CT images (n=51) or low-dose CT images from PET/CT (n=33), primary tumor sites, contralateral non-tumor sites, and ipsilateral and contralateral oropharyngeal soft tissues were manually segmented on axial CT images using CAVASS software. Morphologic, intensity-based, and texture-based features (680 in total) were then extracted for each tissue and compared for discriminative capability between the two patient cohorts using unpaired t testing with p<0.01 considered as significant. Results 69 (out of 680) radiomic features on baseline CT scans were significantly different between patients with recurrence vs. without recurrence after TORS. Volume of oropharyngeal tissue was greater in patients who experienced recurrence, on both tumor (ipsilateral, 36.22±9.28cc vs. 25.39±11.43cc, p<0.0001) and non-tumor (contralateral, 33.62±7.24cc vs. 24.73±11.83cc, p<0.0001) sides. Additionally, the mean CT intensity value of ipsilateral oropharyngeal tissue in the recurrence group was significantly higher than that in the control group (47±16 vs. 34±26 Hounsfield units, p <0.01). Several texture-based features also differed between the groups. Conclusion Radiomic features can distinguish between HPV+ OPSCCs that do and do not recur after TORS. Oropharyngeal tissue volume seems to be greater in patients with recurrence. Further analysis and validation on external cohorts are needed to establish clinical utility.