Artificial Intelligence in Medicine: Fully Automated Estimation of Gleason Grade Using AI?

It’s hard to imagine the imagination of science fiction writers. While many of their literary works are based in our current world, like Michael Crichton’s novel The Andromeda Strain, science fiction works are not bound by past, present or future. In fact, the majority envision scientific or technological advances that don’t yet exist. As a blog from high-tech company Micron puts it, “Some of the most outlandish scenarios imagined by writers of films, TV shows, and books have come true, and they were actually inspired by science fiction.” It presents 11 examples that were anticipated by sci-fi writers, including self-driving cars, 3-D food printing, drones, ear buds and more.

Predicting inventions and their achievements is not just the realm of fiction. In 2016, two Harvard academics teamed up to announce that “machine learning will displace much of the work of radiologists and anatomical pathologists.”^[i] It didn’t take long to start. By 2020, a multicenter team developed and tested a Deep Learning (DL) system to assign a Gleason grade to prostate cancer (PCa) biopsy specimens. They found that their program demonstrated “higher proficiency than general pathologists at Gleason grading prostate needle core biopsy specimens…”^[ii]

The pace of generating such programs is accelerating. In 2022, a team of Spanish researchers tackled the Herculean chore of employing DL to “… simultaneously perform detection, segmentation, and Gleason Grade estimation from [multiparametric MRIs] to a state-of-the-art performance level.”^[iii] In publishing their work, they note that Computer-Aided Diagnosis (CAD) has already been acknowledged as a way to add efficiency and objectivity to the human task of interpreting mpMRI scans. After all, reading and reporting on scans is time-consuming, and dependent on the expertise of the reader so results can vary, so carefully designed and trained algorithmic programs can help busy radiologists by detecting suspicious lesions.

The Spanish team’s project goes further. They based their product on “two main MRI datasets integrating T2, DW, b-value, and ADC maps in both of them” as well as two PCa datasets (ProstateX and IVO). Their image-based lesion identification included prostate zonal segmentation and classification by grade, which was then compared against “fusion and transperineal template biopsies,” the latter of which is considered the pre-operative gold standard to evaluate PCa extent.

How well did their DL system perform? In terms of detecting clinically significant PCa (that is, Gleason grade group ≥ 2), their results were excellent. For the IVO dataset, their program outperformed PI-RADS for high risk PCa. They also found great agreement in zonal segmentation between their program and that of an expert radiologist. While not yet at a point where the authors can claim to have perfected automated image-based diagnosis, they can justly pat themselves on the back for having created “the first algorithm, to the best of our knowledge, that automatically contours the prostate into its zones, performs well at lesion detection and Gleason Grade prediction (identifying lesions of a given grade or higher), and segments such lesions albeit with a moderate overlapping. The model outperformed expert radiologists with extensive MRI experience …”

Our team at Sperling Prostate Center extends congratulations to Pellicer-Valero, et al. for their thorough work. They have made a significant stride in the contribution of DL as a promising working partner with human radiologists who specialize in prostate cancer.

NOTE: This content is solely for purposes of information and does not substitute for diagnostic or medical advice. Talk to your doctor if you are experiencing pelvic pain, or have any other health concerns or questions of a personal medical nature.

[i] Obermeyer Z, Emanuel EJ. Predicting the Future – Big Data, Machine Learning, and Clinical Medicine. N Engl J Med. 2016 Sep 29;375(13):1216-9.
[ii] Nagpal K, Foote D, Tan F, et al. Development and Validation of a Deep Learning Algorithm for Gleason Grading of Prostate Cancer From Biopsy Specimens. JAMA Oncol. 2020;6(9):1372–1380.
[iii] Pellicer-Valero OJ, Marenco Jiménez JL, Gonzalez-Perez V, Casanova Ramón-Borja JL et al. Deep learning for fully automatic detection, segmentation, and Gleason grade estimation of prostate cancer in multiparametric magnetic resonance images. Sci Rep. 2022 Feb 22;12(1):2975.