Gynecological Cancer

SS 24 - GYN 1

170 - An MRI-Based Radiomic Signature for Disease-Free Survival?in Locally Advanced Cervical Cancer

Tuesday, October 23
1:40 PM - 1:50 PM
Location: Room 007 A/B

An MRI-Based Radiomic Signature for Disease-Free Survival in Locally Advanced Cervical Cancer
K. Han1,2, M. Welch1, J. Weiss3, M. Pintilie3, T. W. Fyles1,4, and M. Milosevic1,2; 1Radiation Medicine Program, Princess Margaret Cancer Centre, University Health Network, Toronto, ON, Canada, 2Radiation Medicine Program, University Health Network and Princess Margaret Cancer Centre, Toronto, ON, Canada, 3Department of Biostatistics, University Health Network and Princess Margaret Cancer Centre, Toronto, ON, Canada, 4Department of Radiation Oncology, University of Toronto, Toronto, ON, Canada

Purpose/Objective(s): Radiomics is an emerging area of discovery where quantitative imaging features are extracted from routine imaging, and mined to identify potential prognostic imaging phenotypes. The majority of studies have focused on CT images and non-cervical cancer. We hypothesized that MR imaging features can prognosticate disease-free survival (DFS) in locally advanced cervical cancer, and aimed to develop a MR-based radiomic signature for DFS.

Materials/Methods: The study comprised a discovery dataset of 80 patients and an independent validation dataset of 81 patients with FIGO stage IB-IVA cervical cancer treated with definitive chemoradiation between 2005-2014. Disease status was recorded prospectively. The primary end-point was DFS, defined as freedom from relapse or death measured from the date of diagnosis. A single observer retrospectively contoured the primary tumor on T2-weighted pre-treatment MRI. Radiomic features of the tumor were extracted using an open source code package. The stability of each feature was determined by calculating the Pearson correlation coefficients of radiomic features extracted from each of the following settings: (1) test-retest MR images acquired 15 minutes apart, (2) diagnostic and simulation MR images acquired < 10 days apart, and (3) tumor delineation by 2 observers (inter-observer variability). Less stable features with an average Pearson correlation coefficient of < 0.6 were excluded. A radiomic signature for DFS was built using the discovery cohort, with minimal redundancy maximum relevancy feature selection method and 10-fold cross validation. The radiomic signature was dichotomized by its median value into high- and low-risk groups. Uni- and multi-variable Cox regression analyses were performed to evaluate the performance of the radiomic signature in both datasets using the concordance index (CI).

Results: A radiomic signature comprising two features (based on shape and wavelet) was prognostic for DFS in the discovery cohort (HR 2.58, CI 0.65, p = 0.007). The 5-year DFS rates of patients in the high- vs. low-risk radiomic groups were 73 vs 42%, respectively (p = 0.005). Tumor volume, and a clinical model with stage and nodal status were moderately prognostic for DFS in the discovery cohort (CI = 0.60, p = 0.03; and CI 0.64, p = 0.08, respectively). The radiomic signature remained independently associated with DFS when added to tumor volume (HR for radiomic signature 2.29, p = 0.038, model CI 0.65), or the clinical model (HR 2.27, p = 0.02, model CI 0.69) in the discovery cohort. The radiomic signature was also prognostic for DFS in the independent validation cohort, both on univariable analysis (HR 2.39, p = 0.044, CI 0.63), and also when added to a clinical model with stage and nodal status (HR 2.65, p = 0.039, model CI 0.70).

Conclusion: A radiomic signature can be used as a prognostic biomarker for DFS following chemoradiation in patients with locally advanced cervical cancer. Further external validation is planned.

Author Disclosure: K. Han: None. M. Welch: None. M. Pintilie: None. T.W. Fyles: Independent Contractor; University Health Network. M. Milosevic: None.

Kathy Han, MD, MS

Disclosure:
Employment
Princess Margaret Cancer Center: Radiation Oncologist: Employee

Presentation(s):

Send Email for Kathy Han


Assets

170 - An MRI-Based Radiomic Signature for Disease-Free Survival?in Locally Advanced Cervical Cancer



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for An MRI-Based Radiomic Signature for Disease-Free Survival?in Locally Advanced Cervical Cancer