(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: To identify the true pathology before radical prostatectomy, by predicting the lymph node metastases from the radiomic characteristics extracted from preoperative FDG-PET/CT images.
Methods: FDG-PET/CT images were acquired as a staging procedure prior to prostatectomy in 160 patients diagnosed with prostate cancer with a Gleason sum of =8 at biopsy. Only 30% of patients have metastatic lymph nodes. A total of 1015 radiomic features extracted with Pyradiomics was used for the analysis. To reduce the dimensionality of the model and focus only on features carrying useful information, three layers of filtration were applied to the radiomic features: I) modification of the segmentation, II) multicollinearity analysis, and III) mRMR (minimum Redundancy Maximum Relevance). A random forest classifier is used to predict the presence of lymph node metastases.
Results: After filtration, a subset of 331 radiomic characteristics was selected. The accuracy of the model was 74.5%, of which 14% had lymph node metastasis, and the AUC is 63.5%. This corresponds to an increase in precision of 6% and an increase in AUC of 12% compared to a model trained on all features. The sensitivity and specificity are 33.3% and 93.8% respectively. With only 10% of the radiomic markers, a loss of precision and an AUC of 6% is observed. These results show a slight improvement compared to published prediction using only the intraprostatic FDG uptake to characterize the association between imaging and recurrence of cancer. An accuracy of 26% was obtained, of which 39% had lymph node metastasis.
Conclusion: Extensive analysis of radiomic features could improve the accuracy of FDG-PET/CT to predict lymph node metastasis before radical prostatectomy. Ultimately, the algorithm will better predict the risk of recurrent prostate cancer and help improve methods and choice of treatment.
Keywords
Not Applicable / None Entered.
Taxonomy
IM/TH- Image Analysis (Single Modality or Multi-Modality): Imaging biomarkers and radiomics
Contact Email
