Prediction of Adverse Pathology at Radical Prostatectomy in Grade Group 2 and 3 Prostate Biopsies Using Machine...

Area of Science:

• Computational pathology
• Digital pathology
• Machine learning in oncology

Background:

• Clinical need exists for improved prostate cancer outcome prediction.
• Machine learning (ML) offers a promising strategy for pathologic diagnosis and prognosis.
• Current methods require enhancement for predicting adverse outcomes.

Purpose of the Study:

• To develop and evaluate an ML algorithm for predicting adverse outcomes in prostate cancer.
• Utilize whole-slide images (WSIs) of prostate biopsies with Grade Group (GG) 2 or 3 disease.
• Assess ML algorithm performance against clinical risk assessment scores.

Main Methods:

• Retrospective review of prostate biopsies and corresponding radical prostatectomy (RP) data.
• Development of an ML pipeline involving image preprocessing, feature extraction (VGG-16), and classification (Extreme Gradient Boosting).
• Input data consisted of WSIs from hematoxylin and eosin-stained core needle biopsies.

Main Results:

• The ML model achieved an area under the receiver operating characteristic curve of 0.72 for the entire cohort.
• Performance varied by Grade Group: 0.65 for GG 2 and 0.89 for GG 3 patients.
• ML pipeline performance was comparable to the CAPRA clinical risk assessment score.

Conclusions:

• ML algorithms show potential in utilizing WSIs of prostate needle core biopsies.
• This approach can effectively estimate clinically relevant prostate cancer outcomes.
• WSIs combined with ML offer a novel tool for prostate cancer prognosis.