Article Text

Download PDFPDF

262 MRI based tumour radiomics in carcinoma cervix: a potential predictor of response to chemoradiation
  1. S Chopra1,
  2. P Nayak2,
  3. Y Gupta1,
  4. J Swamidas1,
  5. JS Goda1,
  6. S Panda1,
  7. L Gurram1,
  8. S Rath3,
  9. P Popat4,
  10. U Mahantshetty2,
  11. S Gupta3 and
  12. JP Agarwal2
  1. 1ACTREC, Tata Memorial Centre, Radiation Oncology, Navi Mumbai, India
  2. 2Tata Memorial Hospital, Tata Memorial Centre, Radiation Oncology, Mumbai, India
  3. 3Tata Memorial Hospital, Tata Memorial Centre, Medical Oncology, India
  4. 4Tata Memorial Hospital, Tata Memorial Centre, Radiology, India


Introduction/Background*To determine if baseline T2weighted (T2W) MRI texture parameters can predict response to chemoradiation in cervical carcinoma.

Methodology Seventy-four patients of locally advanced carcinoma cervix treated with definitive chemoradiation (45Gy in 25 fractions and weekly cisplatin (40 mg/m2) and image guided brachytherapy between 2017 and 2019 were included. Gross tumour volume (GTV) and high risk clinical target volume (HRCTV) was delineated on T2W MRI at baseline and at brachytherapy using the texture analysis software. Tumour regression >75% was considered as a surrogate of good response. Multiple tumour slices were sampled and first order statistics were applied to produce 6 texture parameters, namely mean intensity, standard deviation, entropy, mean positive pixels, skewness, and kurtosis. After using fine,moderate, and coarse anatomical filters a total of 36 texture variables were generated.

Clinical variables namely histological subtype, tumour grade, FIGO 2009 stage and nodal status were documented. These clinical variables along with texture parameters were compared with treatment response using the Mann-Whitney U test. Lasso regression was used to select texture parameters that best correlated with treatment response. These were used to develop support vector machine (SVM) models which were validated using 10-fold cross validation. The most parsimonious model was described in terms of area under curve (AUC) and metrics of diagnostic accuracy.

Result(s)*The median age was 50 years (range 34 to 65). Overall, 63 (85%) had squamous cell carcinoma. Half of the included patients (37/74) had positive pelvic or para-aortic lymph nodes. As per FIGO 2018 criteria 23 (31%) and 51 (69%) patients were stage II and III, respectively. Good response was seen in 10/74 patients. None of the clinical variables discriminated between response. However, high mean and skewness, and low entropy and kurtosis did significantly correlate with poor response. Using 15 selected features, the best SVM model had an AUC of 0.85 and accurately classified 86.5% cases, with sensitivity and specificity of 55% and 93%, respectively.

Conclusion*Tumour radiomics can be utilized to predict response to chemoradiation whereas baseline clinical parameters do not predict for response. Baseline texture analysis can be an important tool to predict response to chemoradiation.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.