Application of Artificial Intelligence/Machine Vision & Learning for the Development of a Live Single-cell Phenotypic...

Area of Science:

• Computational biology and bioinformatics
• Oncology and cancer research
• Medical imaging and machine vision

Background:

• Current prostate cancer risk stratification relies on pathology reports, which may not fully capture tumor aggressiveness.
• Tumor heterogeneity and molecular complexity present challenges in accurate cancer assessment.
• There is a need for advanced tools to complement existing methods for predicting adverse pathology.

Purpose of the Study:

• To assess the utility of machine vision (MV) and machine learning (ML) for a single-cell phenotypic platform.
• To correlate live and fixed cell biomarkers with prostate tumor aggressiveness and risk stratification.
• To develop a novel prognostic and risk-stratification tool for prostate cancer patients.

Main Methods:

• 100 fresh prostate cancer samples were dissociated into single-cell suspensions and analyzed via live-cell microscopy.
• Phenotypic biomarkers were quantified using MV software and ML algorithms, including random forest (RF) models.
• ML algorithms were trained on 70% of samples and validated on a blinded 30% test set, using ROC curve analysis.

Main Results:

• MV and ML algorithms identified predictive biomarkers for adverse pathology in radical prostatectomy specimens.
• Developed algorithms achieved >0.85 sensitivity and specificity, with an AUC >0.85 for predicting adverse pathology.
• A novel risk stratification test was created, distinguishing patients based on surgical adverse pathology features.

Conclusions:

• Single-cell phenotypic biomarkers, when analyzed with MV and ML, are informative for predicting post-surgical adverse pathologies.
• AI-based ML approaches offer powerful tools to enhance current prostate cancer risk stratification methods.
• The developed phenotypic test represents a novel, powerful prognostic tool leveraging advanced MV and ML techniques.