Photo Credit: David Gyung

The following is a summary of “Clinical application of machine learning models in patients with prostate cancer before prostatectomy,” published in the February 2024 issue of Oncology by Guerra et al.

This retrospective observational study aimed to construct machine learning predictive models to assess surgical risk related to extracapsular extension (ECE) in prostate cancer (PCa) patients before radical prostatectomy. Two independent datasets were analyzed: one comprising 139 participants from a single institution (used for training) and another consisting of 55 patients from 15 different institutions (utilized for external validation), all treated with Robotic Assisted Radical Prostatectomy (RARP). Five machine learning models, incorporating various combinations of clinical, semantic (interpreted by a radiologist), and radiomics features derived from T2W-MRI images, were developed to predict extracapsular extension in prostatectomy specimens (pECE+). 

Decision curve analysis (DCA) plots were employed to evaluate the models’ net benefit in assigning patients to prostatectomy with either non-nerve-sparing surgery (NNSS) or nerve-sparing surgery (NSS) based on the predicted ECE status. The rankings of models derived from DCA were compared with those derived from the receiver operating characteristic (ROC) area under the curve (AUC). Results indicated that the model integrating clinical, semantic, and radiomics features demonstrated the highest net benefit values across relevant threshold probabilities in the training data, with similar performance observed in the external validation set. 

However, the model ranking based on AUC differed between the discovery groups and favored the model utilizing only clinical and semantic features. In conclusion, the combined model incorporating clinical, semantic, and radiomic features shows promise in predicting pECE+ in PCa patients, offering a positive net benefit when determining between prostatectomy with NNSS or NSS.

 Source: cancerimagingjournal.biomedcentral.com/articles/10.1186/s40644-024-00666-y