Author: Almudena Fuster
Date: 08th Jun 2022
Prostate cancer (PCa) is the second most frequent malignancy in men worldwide, accounting for 1,414,259 new cases and causing around 3.8% of all deaths from cancer in men in 20201. PCa diagnosis is not ideal. It usually requires a transrectal ultrasound-guided (TRUS) biopsy, a painful procedure not always exempted from certain side-effects2 and that has been reported to lead to sampling errors, overdetection of indolent PCa, and misclassification3.
Multiparametric magnetic resonance imaging (mpMRI) provides a better visualization of the entire prostate gland in a completely non-invasive way. This test can help physicians rule out the need for a biopsy and, if needed, it provides valuable information so that biopsies are taken from the most suspicious areas and not just randomly.
At present, there is mounting evidence that mpMRI prior to biopsies provides a more accurate diagnosis4,5, being endorsed by several clinical practice guidelines6,7. However, the population-wide adoption of mpMRI still needs to overcome some hurdles. On the one hand, since it may clearly change the PCa pathway, a sharp increase in the number of men who will undergo prostate MRI is expected. Providing good image quality and diagnostic accuracy while meeting the demands of the expected high workload will be challenging. On the other hand, from a clinical point of view, the greatest concern arises from its highly variable accuracy across institutions and among individual radiologists8. Despite the implementation of a standardized reporting system such as the Prostate Imaging Reporting and Data System (PI-RADS)9, there is still room for improvement in mpMRI reporting. Finally, and as discussed by the American Urologists Association (AUA),“there are several significant impediments to the adoption of mpMRI as a stand-alone, population-based screening strategy”6, so it seems that there is still a long way to go before MRI scales up as a standard-of-care screening tool, as mammography did in breast cancer. However…
Could artificial intelligence (AI) help to create more reliable and faster MRI scans speeding up this process?
AI is a rapidly emerging technology and has gained massive interest in medical imaging research, mainly in a preclinical setting10.
Several studies have evaluated the performance of computer-aided diagnosis (CAD) coupled with radiomics and machine learning to diagnose PCa from clinical images with promising results11. Recently, the group led by Nickolas Papanikolaou at the Champalimaud Foundation in Lisbon, Portugal, has provided new insights into the prediction of PCa disease aggressiveness using MRI radiomics, a methodology that enables to transform medical images into high dimension mineable data12. Thus, based on the hypothesis that tumor tissue characteristics can be quantified by radiomic features extracted from bi-parametric MRI (a shorter type of MRI scan that uses the same device to image the prostate to detect signs of cancer), the investigators developed different supervised machine learning models to predict biological aggressiveness. Their results demonstrated that MRI-based radiomic features allowed the identification of clinically significant PCa. Also, they observed that radiomic features extracted from the whole prostate gland produced better models than those extracted from specific regions, highlighting the relevance of those areas surrounding the tumor lesions as important sources of information.
Interestingly, AI does not only have the potential to improve diagnosis but can also be useful in metastasis detection and prediction of response to treatment. Indeed, those are some of the aims of ProCAncer-I, a project led by Dr. Papanikolau among others, that brings together 20 partners, including PCa centers of reference, world leaders in AI, and innovative small and medium-sized enterprises (https://www.procancer-i.eu/) and that aspires to create the largest interoperable, high-quality mpMRI dataset worldwide/globally (currently including >17,000 cases). Quibim is proud to lead tasks within key work packages of this ambitious initiative, including image and data annotation-related activities, and the implementation of monitoring, logging, and retraining of AI models.
Although AI can certainly help in mitigating some of the already discussed pitfalls related to biopsy and/or to conventional mpMRI, at present there is insufficient evidence to suggest the clinical deployment of AI intelligence algorithms. However, AI will surely change the landscape in clinical practice, by replacing certain tasks/procedures while complementing overall decision-making.