An integrated radiology-pathology machine learning classifier for outcome prediction following radical prostatectomy: Preliminary findings

Amogh Hiremath; Germán Corredor; Lin Li; Patrick Leo; Cristina Magi-Galluzzi; Robin Elliott; Andrei Purysko; Rakesh Shiradkar; Anant Madabhushi

doi:10.1016/j.heliyon.2024.e29602

An integrated radiology-pathology machine learning classifier for outcome prediction following radical prostatectomy: Preliminary findings

Heliyon. 2024 Apr 15;10(8):e29602. doi: 10.1016/j.heliyon.2024.e29602. eCollection 2024 Apr 30.

Authors

Amogh Hiremath¹, Germán Corredor², Lin Li³, Patrick Leo³, Cristina Magi-Galluzzi⁴, Robin Elliott⁵, Andrei Purysko⁶, Rakesh Shiradkar², Anant Madabhushi^{2

7}

Affiliations

¹ Picture Health, Cleveland, OH, USA.
² Department of Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, GA, USA.
³ Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, USA.
⁴ Pathology and Laboratory Medicine Institute, Cleveland Clinic, Cleveland, OH, USA.
⁵ Department of Pathology, University Hospitals Cleveland Medical Center, Cleveland, OH, USA.
⁶ Department of Radiology and Nuclear Medicine, Cleveland Clinic, Cleveland, OH, USA.
⁷ Atlanta Veterans Administration Medical Center, Atlanta, GA, USA.

Abstract

Objectives: To evaluate the added benefit of integrating features from pre-treatment MRI (radiomics) and digitized post-surgical pathology slides (pathomics) in prostate cancer (PCa) patients for prognosticating outcomes post radical-prostatectomy (RP) including a) rising prostate specific antigen (PSA), and b) extraprostatic-extension (EPE).

Methods: Multi-institutional data (N = 58) of PCa patients who underwent pre-treatment 3-T MRI prior to RP were included in this retrospective study. Radiomic and pathomic features were extracted from PCa regions on MRI and RP specimens delineated by expert clinicians. On training set (D1, N = 44), Cox Proportional-Hazards models M_R, M_P and M_RaP were trained using radiomics, pathomics, and their combination, respectively, to prognosticate rising PSA (PSA > 0.03 ng/mL). Top features from M_RaP were used to train a model to predict EPE on D1 and test on external dataset (D2, N = 14). C-index, Kalplan-Meier curves were used for survival analysis, and area under ROC (AUC) was used for EPE. M_RaP was compared with the existing post-treatment risk-calculator, CAPRA (M_C).

Results: Patients had median follow-up of 34 months. M_RaP (c-index = 0.685 ± 0.05) significantly outperformed M_R (c-index = 0.646 ± 0.05), M_P (c-index = 0.631 ± 0.06) and M_C (c-index = 0.601 ± 0.071) (p < 0.0001). Cross-validated Kaplan-Meier curves showed significant separation among risk groups for rising PSA for M_RaP (p < 0.005, Hazard Ratio (HR) = 11.36) as compared to M_R (p = 0.64, HR = 1.33), M_P (p = 0.19, HR = 2.82) and M_C (p = 0.10, HR = 3.05). Integrated radio-pathomic model M_RaP (AUC = 0.80) outperformed M_R (AUC = 0.57) and M_P (AUC = 0.76) in predicting EPE on external-data (D2).

Conclusions: Results from this preliminary study suggest that a combination of radiomic and pathomic features can better predict post-surgical outcomes (rising PSA and EPE) compared to either of them individually as well as extant prognostic nomogram (CAPRA).

Keywords: Histopathology; MRI; Machine learning; Prognosis; Prostate cancer.