A machine learning approach to predict progression on active surveillance for prostate cancer

Comparing Active Surveillance and Watchful Waiting With Radical Treatment Using Machine Learning Models Among Patients With Prostate Cancer

PURPOSE

In 2021, 59.6% of low-risk patients with prostate cancer were under active surveillance (AS) as their first course of treatment. However, few studies have investigated AS and watchful waiting (WW) separately. The objectives of this study were to develop and validate a population-level machine learning model for distinguishing AS and WW in the conservative treatment group, and to investigate initial cancer management trends from 2004 to 2017 and the risk of chronic diseases among patients with prostate cancer with different treatment modalities.

METHODS

In a cohort of 18,134 patients with prostate adenocarcinoma diagnosed between 2004 and 2017, 1,926 patients with available AS/WW information were analyzed using machine learning algorithms with 10-fold cross-validation. Models were evaluated using performance metrics and Brier score. Cox proportional hazard models were used to estimate hazard ratios for chronic disease risk.

RESULTS

Logistic regression models achieved a test area under the receiver operating curve of 0.73, F-score of 0.79, accuracy of 0.71, and Brier score of 0.29, demonstrating good calibration, precision, and recall values. We noted a sharp increase in AS use between 2004 and 2016 among patients with low-risk prostate cancer and a moderate increase among intermediate-risk patients between 2008 and 2017. Compared with the AS group, radical treatment was associated with a lower risk of prostate cancer-specific mortality but higher risks of Alzheimer disease, anemia, glaucoma, hyperlipidemia, and hypertension.

CONCLUSION

A machine learning approach accurately distinguished AS and WW groups in conservative treatment in this decision analytical model study. Our results provide insight into the necessity to separate AS and WW in population-based studies.

Stability of Multi-Parametric Prostate MRI Radiomic Features to Variations in Segmentation

Stability analysis remains a fundamental step in developing a successful imaging biomarker to personalize oncological strategies. This study proposes an in silico contour generation method for simulating segmentation variations to identify stable radiomic features. Ground-truth annotation provided for the whole prostate gland on the multi-parametric MRI sequences (T2w, ADC, and SUB-DCE) were perturbed to mimic segmentation differences observed among human annotators. In total, we generated 15 synthetic contours for a given image-segmentation pair. One thousand two hundred twenty-four unfiltered/filtered radiomic features were extracted applying Pyradiomics, followed by stability assessment using ICC(1,1). Stable features identified in the internal population were then compared with an external population to discover and report robust features. Finally, we also investigated the impact of a wide range of filtering strategies on the stability of features. The percentage of unfiltered (filtered) features that remained robust subjected to segmentation variations were T2w-36% (81%), ADC-36% (94%), and SUB-43% (93%). Our findings suggest that segmentation variations can significantly impact radiomic feature stability but can be mitigated by including pre-filtering strategies as part of the feature extraction pipeline.

Development, comparison, and validation of four intelligent, practical machine learning models for patients with prostate-specific antigen in the gray zone

Purpose

Machine learning prediction models based on LogisticRegression, XGBoost, GaussianNB, and LGBMClassifier for patients in the prostate-specific antigen gray zone are to be developed and compared, identifying valuable predictors. Predictive models are to be integrated into actual clinical decisions.

Methods

Patient information was collected from December 01, 2014 to December 01, 2022 from the Department of Urology, The First Affiliated Hospital of Nanchang University. Patients with a pathological diagnosis of prostate hyperplasia or prostate cancer (any PCa) and having a prostate-specific antigen (PSA) level of 4-10 ng/mL before prostate puncture were included in the initial information collection. Eventually, 756 patients were selected. Age, total prostate-specific antigen (tPSA), free prostate-specific antigen (fPSA), fPSA/tPSA, prostate volume (PV), prostate-specific antigen density (PSAD), (fPSA/tPSA)/PSAD, and the prostate MRI results of these patients were recorded. After univariate and multivariate logistic analyses, statistically significant predictors were screened to build and compare machine learning models based on LogisticRegression, XGBoost, GaussianNB, and LGBMClassifier to determine more valuable predictors.

Results

Machine learning prediction models based on LogisticRegression, XGBoost, GaussianNB, and LGBMClassifier exhibit higher predictive power than individual metrics. The area under the curve (AUC) (95% CI), accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score of the LogisticRegression machine learning prediction model were 0.932 (0.881-0.983), 0.792, 0.824, 0.919, 0.652, 0.920, and 0.728, respectively; of the XGBoost machine learning prediction model were 0.813 (0.723-0.904), 0.771, 0.800, 0.768, 0.737, 0.793 and 0.767, respectively; of the GaussianNB machine learning prediction model were 0.902 (0.843-0.962), 0.813, 0.875, 0.819, 0.600, 0.909, and 0.712, respectively; and of the LGBMClassifier machine learning prediction model were 0.886 (0.809-0.963), 0.833, 0.882, 0.806, 0.725, 0.911, and 0.796, respectively. The LogisticRegression machine learning prediction model has the highest AUC among all prediction models, and the difference between the AUC of the LogisticRegression prediction model and those of XGBoost, GaussianNB, and LGBMClassifier is statistically significant (p < 0.001).

Conclusion

Machine learning prediction models based on LogisticRegression, XGBoost, GaussianNB, and LGBMClassifier algorithms exhibit superior predictability for patients in the PSA gray area, with the LogisticRegression model yielding the best prediction. The aforementioned predictive models can be used for actual clinical decision-making.​.

Developing machine learning algorithms for dynamic estimation of progression during active surveillance for prostate cancer

Active Surveillance (AS) for prostate cancer is a management option that continually monitors early disease and considers intervention if progression occurs. A robust method to incorporate “live” updates of progression risk during follow-up has hitherto been lacking. To address this, we developed a deep learning-based individualised longitudinal survival model using Dynamic-DeepHit-Lite (DDHL) that learns data-driven distribution of time-to-event outcomes. Further refining outputs, we used a reinforcement learning approach (Actor-Critic) for temporal predictive clustering (AC-TPC) to discover groups with similar time-to-event outcomes to support clinical utility. We applied these methods to data from 585 men on AS with longitudinal and comprehensive follow-up (median 4.4 years). Time-dependent C-indices and Brier scores were calculated and compared to Cox regression and landmarking methods. Both Cox and DDHL models including only baseline variables showed comparable C-indices but the DDHL model performance improved with additional follow-up data. With 3 years of data collection and 3 years follow-up the DDHL model had a C-index of 0.79 (±0.11) compared to 0.70 (±0.15) for landmarking Cox and 0.67 (±0.09) for baseline Cox only. Model calibration was good across all models tested. The AC-TPC method further discovered 4 distinct outcome-related temporal clusters with distinct progression trajectories. Those in the lowest risk cluster had negligible progression risk while those in the highest cluster had a 50% risk of progression by 5 years. In summary, we report a novel machine learning approach to inform personalised follow-up during active surveillance which improves predictive power with increasing data input over time.

Machine Learning in Prostate MRI for Prostate Cancer: Current Status and Future Opportunities

Advances in our understanding of the role of magnetic resonance imaging (MRI) for the detection of prostate cancer have enabled its integration into clinical routines in the past two decades. The Prostate Imaging Reporting and Data System (PI-RADS) is an established imaging-based scoring system that scores the probability of clinically significant prostate cancer on MRI to guide management. Image fusion technology allows one to combine the superior soft tissue contrast resolution of MRI, with real-time anatomical depiction using ultrasound or computed tomography. This allows the accurate mapping of prostate cancer for targeted biopsy and treatment. Machine learning provides vast opportunities for automated organ and lesion depiction that could increase the reproducibility of PI-RADS categorisation, and improve co-registration across imaging modalities to enhance diagnostic and treatment methods that can then be individualised based on clinical risk of malignancy. In this article, we provide a comprehensive and contemporary review of advancements, and share insights into new opportunities in this field.

Predicting survival after radical prostatectomy: Variation of machine learning performance by race

BACKGROUND

Robust prediction of survival can facilitate clinical decision-making and patient counselling. Non-Caucasian males are underrepresented in most prostate cancer databases. We evaluated the variation in performance of a machine learning (ML) algorithm trained to predict survival after radical prostatectomy in race subgroups.

METHODS

We used the National Cancer Database (NCDB) to identify patients undergoing radical prostatectomy between 2004 and 2016. We grouped patients by race into Caucasian, African-American, or non-Caucasian, non-African-American (NCNAA) subgroups. We trained an Extreme Gradient Boosting (XGBoost) classifier to predict 5-year survival in different training samples: naturally race-imbalanced, race-specific, and synthetically race-balanced. We evaluated performance in the test sets.

RESULTS

A total of 68,630 patients met inclusion criteria. Of these, 57,635 (84%) were Caucasian, 8173 (12%) were African-American, and 2822 (4%) were NCNAA. For the classifier trained in the naturally race-imbalanced sample, the F1 scores were 0.514 (95% confidence interval: 0.513-0.511), 0.511 (0.511-0.512), 0.545 (0.541-0.548), and 0.378 (0.378-0.389) in the race-imbalanced, Caucasian, African-American, and NCNAA test samples, respectively. For all race subgroups, the F1 scores of classifiers trained in the race-specific or synthetically race-balanced samples demonstrated similar performance compared to training in the naturally race-imbalanced sample.

CONCLUSIONS

A ML algorithm trained using NCDB data to predict survival after radical prostatectomy demonstrates variation in performance by race, regardless of whether the algorithm is trained in a naturally race-imbalanced, race-specific, or synthetically race-balanced sample. These results emphasize the importance of thoroughly evaluating ML algorithms in race subgroups before clinical deployment to avoid potential disparities in care.