PI-CAI: Prostate Cancer MRI Scans for AI Validation

From 2012 to 2021, 1500 anonymized prostate biparametric MRI scans were collected from 1476 patients across three medical centers in the Netherlands. This dataset is the public training and development set for the PI-CAI challenge, which aims to validate the diagnostic performance of AI algorithms against radiologists for detecting clinically significant prostate cancer. The reference standard is histopathology and follow-up data of at least three years.

Use Cases

• Training AI models for prostate cancer detection based on biparametric MRI scans.
• Benchmarking algorithm performance against radiologist readings for csPCa diagnosis.
• Validating AI diagnostic tools using a reference standard of histopathology and long-term follow-up.
• Studying the standardization of AI validation in a multi-center, retrospective clinical setting.

Strengths

• Contains 1500 MRI scans from a substantial cohort of 1476 patients.
• Data is clinically validated with a reference standard of histopathology and follow-up of at least three years.
• Collected from three medical centers, suggesting multi-institutional diversity.

Limitations

• Specific column names and data structure are not provided in the source descriptions.
• Metadata such as exact file formats, size, and last update date are unknown.
• The dataset appears focused on a specific clinical challenge, which may limit generalizability to other diagnostic tasks.

Provenance

Source

Radboud University Medical Center, University Medical Center Groningen, Ziekenhuis Groep Twente (The Netherlands).

Collection Method

Retrospective collection of anonymized clinical MRI scans and associated pathology/follow-up data.

Time Range

2012-2021

Geography

The Netherlands