In:
Journal of Clinical Oncology, American Society of Clinical Oncology (ASCO), Vol. 35, No. 15_suppl ( 2017-05-20), p. 4559-4559
Abstract:
4559 Background: After chemotherapy, 〉 50% of patients (pts) with metastatic testicular GCT who undergo retroperitoneal lymph node dissection (RPNLD) for residual masses are found to have fibrosis (F) alone on pathological examination. To minimize overtreatment, better prediction algorithms are needed to identify pts with F who can avoid RPLND. Radiomics uses image processing techniques to extract quantitative textures/features from tumor regions of interest (ROI) to train a classifier that predicts pathological findings. We hypothesized that radiomics may identify pts with a high predicted likelihood of F who may avoid RPLND. Methods: Pts with GCT who had an RPLND for nodal masses 〉 1cm after first line platinum chemotherapy were included. Preoperative contrast enhanced axial CT images of retroperitoneal ROI were manually contoured. 153 radiomics features trained a radial basis function support vector machine classifier to discriminate between viable GCT /Mature Teratoma (T) vs F. Nested ten-fold cross-validation protocol was employed to determine classifier accuracy. Clinical variables and restricted size criteria were used to optimize the classifier. Results: A total of 82 pts with 102 ROI were analyzed (GCT: 21; T: 41; F: 40). The discriminative accuracy of radiomics to identify GCT/T vs F was 72%(±2.2)(AUC: 0.74 (±0.028); positive predictive value: 67% (48-92%); negative predictive value: 74% (62-84%)(p = 0.001)). No major predictive differences were identified when data was restricted by varying maximal axial diameters (AUC range: 0.58(±0.05) - 0.74(±0.03)). Prediction algorithm using clinical variables alone identified an AUC of 0.71 (±0.15). When these variables were added to the radiomic signature, the best performing classifier was identified when axial tumors were limited to diameter 〈 2cm (accuracy: 88.2 (±4.4); AUC: 0.80 (±0.05)(p = 0.02)). Conclusions: A predictive radiomics algorithm had an overall discriminative accuracy of 72% that improved to 88% when combined with clinical details. Further independent validation is required to assess whether radiomics, in conjunction with standard clinical predictors, may allow pts with a high predicted likelihood of F to avoid RPLND.
Type of Medium:
Online Resource
ISSN:
0732-183X
,
1527-7755
DOI:
10.1200/JCO.2017.35.15_suppl.4559
Language:
English
Publisher:
American Society of Clinical Oncology (ASCO)
Publication Date:
2017
detail.hit.zdb_id:
2005181-5