When clinically relevant prostate cancer is found and treated at an early stage before metastasis has occurred, treatments such as prostate cancer surgery and radiation often result in improved survival.[1] Current screening methodologies for prostate cancer include digital rectal examination (DRE) and blood test for prostate-specific antigen (PSA) and common treatments include radical prostatectomy combined with radiation therapy, however, this route is not always a possibility due in part to the complex process required to detect tumors.1 As a result, prostate cancer remains the second leading cause of cancer-related death for men in the US and the sixth most common cancer worldwide.1

New research suggests that prostate cancer screening with magnetic resonance imaging (MRI) may have favorable enough results to be considered as a possible population-based screening method[2]. Although screening by serum PSA levels can reduce prostate cancer-specific mortality, it is unclear whether the benefits outweigh the risks of false-positive results and overdiagnosis of insignificant prostate cancer. False positive results can lead to unnecessary biopsy, overdiagnosis, and over treatment with associated morbidities. PSA screening can also lead to underdiagnosis, with clinically significant cancers being missed.3

The clinical application of MRI in prostate cancer has been gaining an increasingly important role in the management of patients with early-stage prostate cancer including diagnosis in patients with abnormal PSA levels, monitoring of patients on active surveillance, and staging prior to definitive interventions.[3] A recent study suggests that the potential adoption of MRI as an alternative for prostate cancer screening might help improve accuracy, avoid unnecessary biopsies, and identify cancers that would otherwise have been missed.

A potential role for MRI in prostate cancer screening

According to the newest research published in the JAMA Network Open, an MRI-first prostate cancer screening program could prevent one in six deaths.2 Researchers at University College in London, UK created a model to forecast outcomes using different screening methods. The research team utilized short, non-contrast MRI and transrectal ultrasound in comparison with the traditional PSA test. The study was conducted at seven primary care practices and two imaging centers in the UK. Over the course of the prospective, population-based blinded study, a total of 408 men between 50 and 69 years of age participated in the screening program and received all three screening exams.2

Results of each test were independently interpreted and given a score based on a five-point scale of suspicion. At the conclusion of the analysis, the MRI scans were associated with improved detection of clinically significant prostate cancer. Expert commentary on the study offered praise regarding the research and shared excitement at the potential for MRI to be used as a future population-based screening tool.2

Artificial Intelligence-enhanced MRI tools in prostate cancer applications

The radiology Artificial Intelligence (AI) environment is growing rapidly, with an increase in product offerings to support efforts in image acquisition, reading, triage and clinical decision support. Specifically, AI tools such as deep learning and neural networks in MRI are being studied to help radiologists and physicians in a range of clinical settings to help improve their ability to quickly and accurately diagnose prostate cancer.[4],[5],[6] Because when the time to diagnosis and subsequent time to treatment are reduced, patient outcomes improve.

A team of researchers built an AI tool designed to automatically detect and predict the aggressiveness of prostate cancer by analyzing multi-parametric magnetic resonance imaging (mpMRI) on par with experienced readers.4,5,6

Seeking a way to supplement the capabilities and training of less experienced radiologists while also providing for the needs of non-experts in detecting and classifying the aggressiveness of prostate cancer, these researchers chose to focus on developing a deep learning convolutional neural network (CNN) algorithm.4 CNN networks are capable of learning to discriminate between features in an image as well as find nonlinear relationships among complex data.4 These characteristics are quickly making CNN networks and other AI-enabled tools essential elements in the detection and diagnosis of a growing range of cancers.4

Deep learning and AI technology may help MRI overcome human variability in identifying aggressive prostate cancer

Prostate cancer manifests as a range of heterogenous cells and tissue in pattern combinations that require experienced genitourinary radiologists trained to identify and classify them on mpMRI studies.1 Distinguishing between low-, intermediate-, and high-risk tumor tissue can be challenging because it is easily impacted by the level of experience of the radiologist on one hand, and the innate variability that exists in human observation and assessment of images on the other.1

Noninvasive mpMRI is an established and important imaging technique that provides high-contrast and high-resolution visualizations of the prostate and pelvic regions essential for detecting and evaluating aggressiveness in prostate tumors.4,5,6 To date, qualitative or semiquantitative assessment criteria used for interpreting mpMRI findings has often produced a range of varying results due to manual input.4,5,6

Determining if a tumor is malignant or benign and classifying its aggressiveness based on imaging exams requires radiologists to have had the opportunity to train by evaluating thousands of mpMRI scans.4,5,6 As a result, interpretation quality can become variable depending on the training of the reader performing the evaluation.4,5,6 One reason for lack of experience is many healthcare facilities do not have the resources to provide the thousands of scans essential for this specialized training.6

The CNN developed and trained in this study includes a million trainable variables and uses standardized Gleason scores to determine risk for progression, metastasis, and cancer-related death.5 However, the Gleason grading system used to assess low-, intermediate-, or high-risk prostate cancer is itself known to produce variability due to the subjective nature of human observation.1 In addition to reliably capturing the complete range of heterogeneous pattern combinations, the Gleason system has also proven to be an accurate predictor of recurrent disease after the full clinical protocol has been completed.1

Prostate cancer MRI data set trains algorithm for detection, prediction

Researchers internally developed their algorithm by compiling a prostate mpMRI data set from more than 400 patients who received an MRI before undergoing prostatectomy.4,5,6 Patient scans included in the study had a total of nearly 730 tumors that were identified.

Researchers entered the prostate cancer patient data set of mpMRI scans scored by the Gleason system into the deep learning CNN in order to train it using five-fold cross-validation.4,5,6 The algorithm learned to analyze and categorize tumors in a uniform manner and compared the results to the original laboratory tissue specimen.4,5,6

The future potential of MRI and its impact on prostate cancer

The role of MRI in prostate cancer continues to grow in importance and may have the potential to change current screening methods and impact patient outcomes. Its advanced imaging techniques provide clinicians with more accurate prostate cancer identification, as seen in the recent research, and new AI-enhanced MRI tools can offer clinicians increased diagnostic consistency, with the potential to positively impact patient outcomes.   



Read more about our GE Healthcare prostate offerings:


[1] Artificial intelligence at the intersection of pathology and radiology in prostate cancer. Diagnostic Interventional Radiology Journal http://www.dirjournal.org/sayilar/103/buyuk/183-188.pdf Accessed 6/13/2019

[2] Eldred-Evans D, Burak P, Connor MJ, Day E, Evans M, Fiorentino F, Gammon M, Hosking-Jervis F, Klimowska-Nassar N, McGuire W, Padhani AR, Prevost AT, Price D, Sokhi H, Tam H, Winkler M, Ahmed HU. Population-Based Prostate Cancer Screening With Magnetic Resonance Imaging or Ultrasonography: The IP1-PROSTAGRAM Study. JAMA Oncol. 2021 Mar 1;7(3):395-402. doi: 10.1001/jamaoncol.2020.7456. PMID: 33570542; PMCID: PMC7879388.

[3] Wallis CJD, Haider MA, Nam RK. Role of mpMRI of the prostate in screening for prostate cancer. Transl Androl Urol. 2017;6(3):464-471. doi:10.21037/tau.2017.04.31

[4] Joint Prostate Cancer Detection and Gleason Score Prediction in mp-MRI via FocalNet. IEEE Transactions on Medical Imaging https://ieeexplore.ieee.org/document/8653866/metrics#metrics Accessed 6/13/2019

[5] AI can identify, classify prostate cancer on mpMRI. Aunt Minnie https://www.auntminnie.com/index.aspx?sec=sup&sub=aic&pag=dis&ItemID=125223 Accessed 6/13/2019

[6] Artificial intelligence performs as well as experienced radiologists in detecting prostate cancer. UCLA Newsroom http://newsroom.ucla.edu/releases/artificial-intelligence-radiologists-detecting-prostate-cancer Accessed 6/13/2019