Largest Collection of Prostate Cancer Biopsy Images Learned by AI Algorithm to Assist Urologists on Early Diagnosis

A new study was able to consolidate 95,000 datasets of cancer biopsy images. This collection is the largest recorded case of prostate cancer that will utilize the learning process of a specialized artificial intelligence algorithm in diagnosing and grading the said condition in real-world patients.

The research will be presented at the annual meeting of the European Association of Urology (EAU22). The work is part of the wider search for large-scale solutions that use the capability of AI programs for global health advantage, including prognostication, diagnostics, and therapy against prostate cancer.

Prostate Cancer and the State of Urologists

Today, many countries lack pathologists that specialize in urology. Because of this, the scientific community aims to develop AI-powered models to help experts determine prostate cancer's status at an early stage. However, the main issue with this approach is that many facilities have methods to characterize the condition.

For example, the scan images and other samples are prepared differently by AI systems from one clinical laboratory to the other, giving the scientists much harder work than easing the process.

Scholars developed the universal application from the Karolinska Institutet. The team collaborated with experts from Finland's University of Turki, Netherlands' the Radboud University Medical Center, and Google Health to provide an AI model that will go head-to-head in competition with approximately 1,300 separate developers.

The mission is to assemble an algorithm that would be able to grade prostate cancer tumors. The training of the AI software involved 10,000 scan images in the initial phase. Most of the results provided by the developers outmatched the performance of general pathologists and were comparable to the performance of uropathology specialists.

AI Algorithm Developed to Help Provide DIagnosis and Solutions for Prostate Cancer

Karolina Institutet's Department of Medical Epidemiology and Biostatistics expert Kimmo Kartasalo, who also served as lead of the study, explained that grading prostate cancer is the key step to selecting the most appropriate treatment for the patient, but this approach is subjective and varies in between assessments of separate pathologists.

The new AI model could dive further into the assessment, and these opinions would offset standardized grading and minimize the impacts of the lack of available pathologists, Kartasako continued.

Fellow Karolinka specialist Nita Mulliqi and colleagues used 95,000 biopsy image datasets for their model to have a bigger perspective of the condition and to train well against other entries participating in the project. This collection equates to more than three years of study held by a single uropathologist's works.

Most cases in the dataset were gathered from Stockholm over four years, starting in 2012. It also included images from other partner facilities located in Europe and Australia.

With the diverse records presented to the system, the AI is expected to learn and grasp accurate information during the early-stage diagnosis and obtain sufficient solutions from other complexities such as rare cancer types and conditions that mimic the illness, EurekAlert reports.

Check out more news and information on Artificial intelligence in Science Times.