(UroToday.com) The 2023 European Association of Urology (EAU) annual congress held in Milan, Italy between March 10th and 13th, 2023 was host to a prostate cancer abstract session on the role of imaging and PSA density for biopsy indication and tumor staging. Dr. Wietske Luining presented the results of her team’s study validating an 18F-DCFPyL PET radiomics-based machine learning model in intermediate- to high-risk primary prostate cancer.
Machine learning is a branch of artificial intelligence (AI), whereby an algorithm is trained to recognize patterns in a training dataset and is subsequently applied to new, “testing” datasets. This algorithm requires input, which may take the form of imaging data, such as radiomics, and helps classify, identify, or predict outcomes. A critical element of such algorithms/models is external validation, which is required to assess the accuracy and reproducibility of the model.
In a prior study, Dr. Luining and colleagues developed a machine learning-based algorithm utilizing 18F-PSMA-PET/CT radiomics features in patients (n=76) with intermediate- to high-risk prostate cancer scheduled for a robotic radical prostatectomy with an extended pelvic lymph node dissection to predict:
- Lymph node involvement (pN0-1)
- Extraprostatic extension (pT3 or worse)
- High grade pathologic Gleason Score (8 or worse)
- Metastatic disease
As demonstrated in the receiver-operating characteristic (ROC) curve below, the model performed well for the prediction of the outcomes of interest with areas under the curve (AUCs) ranging from 0.76 to 0.86. However, this is to be expected when the algorithm is applied to the training dataset and external validation is required to truly test the accuracy of this predictive algorithm.
In this validation study, the authors included 51 patients with biopsy-proven, intermediate- to high-risk prostate cancer planned for a robotic radical prostatectomy with an extended pelvic lymph node dissection. All patients underwent a pre-operative 18F-DCFPyL-PET/CT prior to the prostatectomy. The lesion of interest was segmented and a total of 480 radiomics features were extracted based on texture, intensity, and morphology. The algorithm was subsequently applied to the testing dataset.
The results of the validation analysis are demonstrated below. For reference, an AUC of 0.50 represents random (i.e. “50-50”) chance. As expected, the algorithm fared much worse in the external validation dataset. The model’s predictive ability was weak for the prediction of lymph node involvement (AUC: 0.57) and extraprostatic extension (AUC: 0.63). However, the model did fare significantly better for prediction of a pathologic high-grade Gleason Score (AUC: 0.84).
Dr. Luining concluded by noting that this machine-learning based algorithm incorporating radiomics features could only reliably predict high-grade pathologic Gleason Score in intermediate- to high-risk prostate cancer patients undergoing a radical prostatectomy. These results underline the need for external and/or multicenter validation of PET radiomics-based machine learning model analyses to assess their reproducibility.
Presented by: Dr. Wietske I. Luining, MD, PhD (candidate), Department of Urology, Amsterdam University Medical Center, Amsterdam, The Netherlands
Written by: Rashid K. Sayyid, MD, MSc – Society of Urologic Oncology (SUO) Clinical Fellow at The University of Toronto, @rksayyid on Twitter during the 2023 European Association of Urology (EAU) Annual Meeting, Milan, IT, Fri, Mar 10 – Mon, Mar 13, 2023.