Post-Prostatectomy Risk Stratification of Biochemical Recurrence Using Transfer Learning-Based Multi-Modal Artificial Intelligence

PURPOSE/OBJECTIVE(S)

For patients undergoing radical prostatectomy for prostate cancer (PCa), accurate risk stratification is essential to guide post-prostatectomy therapeutic decision making. Recently, there has been success in the use of multi-modal artificial intelligence models for men after prostate biopsy to aid in risk stratification. Herein, we trained and tested a TRansfer learning-based multi-modal Artificial InteLligence model (TRAIL) for biochemical recurrence (BCR) risk stratification following radical prostatectomy.

MATERIALS/METHODS

Patients contained within a prospective PCa registry at a single institution were utilized. Digital pathology slides from the diagnostic biopsies prior to radical prostatectomy for patients with clinically localized PCa were scanned at 20x resolution. Features were extracted for the TRAIL model from pathology slides via two transfer learning steps: (1) InceptionResNetv2 that first determines a heatmap of tumor areas, and (2) A ResNet18 that extracts representative features from the high tumor probability areas. Least Absolute Shrinkage and Selection Operator (LASSO) was used for feature selection from the pathology-extracted features. Finally, TRAIL combines the clinical and pathology-extracted features via a classification ensemble model based on weak tree learners to predict 2- and 5-year BCR defined as two consecutive serum PSA levels ≥0.2 ng/mL. TRAIL training was performed on 250 patients and was then locked and applied to the test set of 125 patients. Accuracy and the area under the curve (AUC) were calculated. Comparison to CAPRA-S and to clinical-only features were assessed.

RESULTS

A total of 818 digital whole pathology biopsy slides from 375 patients treated with subsequent radical prostatectomy were included. Surgical margins were positive in 29% of the patients, and 41% had extra-prostatic extension. The median follow-up was 48 months (Range: 1-132 months). The rates of 2-and 5-year BCR were 11% and 18% respectively. A total of 19 digital pathology-driven features were included in TRAIL. Clinical factors included age, ISUPG, Gleason score, PSA, pathological T and N stages, surgical margin involvement, and the presence of extra-prostatic extension. On the testing set, TRAIL achieved a 2-year BCR AUC of 0.76 and accuracy of 0.87, and was superior to CAPRA-S (AUC = 0.57) and clinical-only features (AUC 0.50, accuracy 0.14). For 5-year BCR, TRAIL achieved an AUC of 0.69 and accuracy of 0.78, and performed better than CAPRA-S (AUC = 0.58), and clinical only features (AUC = 0.50, accuracy = 0.23).

CONCLUSION

Through a combination of deep and ensemble learning, TRAIL incorporates clinical and histopathology features, enabling an improved BCR risk stratification post-prostatectomy when compared to the currently used clinicopathologic models. Future work with larger datasets with metastatic events is warranted to further optimize the model for clinical use.

Cross-Comparison Individual Patient Level Analysis of Three Gene Expression Signatures in Localized Prostate in over 50,000 Men

PURPOSE/OBJECTIVE(S)

Risk stratification guides the management of localized prostate cancer. Multiple commercial gene expression biomarkers have been developed to improve estimates of prognosis, however the 22-gene Decipher genomic classifier (22-GC) is the only test with level 1 evidence supporting its use per NCCN guidelines. It is unknown whether other commercial signatures, Oncotype (GPS) or Prolaris (CCP), are sufficiently correlated to negate the differences in evidence supporting these commercial tests. Herein, we aim to perform a cross-comparison of these signatures in a large cohort of patients diagnosed with localized prostate cancer.

MATERIALS/METHODS

Patients diagnosed with localized prostate cancer who underwent whole transcriptome gene expression microarray analysis on their primary tumor biopsy specimen were included. The 22-GC score was calculated by Veracyte using a commercially locked model. Individual genes in each of the GPS and CCP gene signatures were identified, and the gene weights in each signature were retrained for prediction of metastasis in a multi-institutional cohort of 1,574 men with long-term outcome data. This was performed to improve correlation performance of GPS and CCP given only the 22-GC was trained for prediction of metastasis. For each of the three signatures, both continuous and categorical scores were calculated. Linear regression and spearman correlations were calculated both on univariable and multivariable analyses adjusting for age, grade group, PSA, and T-stage.

RESULTS

A total of 50,881 patients were included (15,379 (30.2%) NCCN low-risk, 14,773 (29.0%) favorable intermediate-risk, 15,544 (30.5%) unfavorable intermediate-risk, and 5,185 (10.2%) high/very high-risk) with a median age of 68 years, and a median PSA of 6.2 ng/mL. On linear regression, the GPS model had poor goodness-of-fit to the 22-GC with an R2 of 0.36, as did the CCP model to the 22-GC with an R2 of 0.32. For CCP, the linear sum of the 31-genes was also tested but had inferior performance (R2 0.28) compared to the reoptimized CCP model. Results were similar on multivariable analysis adjusting for age, PSA, clinical stage and grade group. Spearman correlation between the continuous GPS model scores and the 22-GC was moderate at 0.59, as was the correlation between CCP model and the 22-GC of 0.54. CCP is a measure of proliferation, but in 22-GC high-risk patients, the majority (64.1%) of patients had low-average proliferation and only 35.9% had high proliferation, potentially explaining the lack of strong correlation.

CONCLUSION

There is minimal to moderate correlation between the 22-GC and GPS or CCP gene expression signatures tested. Therefore, these tests should not be viewed as interchangeable, and utilization should be based on the level of evidence supporting each gene expression biomarker.