AI Helps Rule Out Unnecessary Biopsies

Let’s start with the hypothetical case of Robert. He considers himself lucky in life because good things have come his way. He has a happy marriage, nice kids, a decent job, and he’s healthy. He follows the principles of the Mediterranean diet, doesn’t smoke, and has an occasional glass of wine. He does cardio workouts at least 3 times per week. He has no known family history of cancer or heart disease. He sees his primary care doc annually, and up till now his PSA has consistently been less than 3.0 ng/mL. He has every reason to feel grateful and not take his good fortune for granted.

Then the unthinkable occurs. His next annual blood test reveals a PSA jump to 4.7 ng/mL. Possible lab error? A repeat PSA in 2 months comes back at 4.8 ng. mL. He’s not feeling lucky when his doc refers him for a needle biopsy.

Not so fast! A PSA alone is not specific for prostate cancer (PCa). It can mean a number of things. Thankfully, there’s a noninvasive additional step before rushing to biopsy. It’s called multiparametric MRI (mpMRI). When done on a powerful 3T magnet, experts from research centers like University College London (UCL) have shown that a pre-biopsy mpMRI can reduce “the need for an upfront tissue biopsy, which is invasive and can lead to complications. As a result, demand for scans has risen sharply.”

However, they caution that “only scans of the highest quality can rule out cancer without the need for a subsequent biopsy, or rule in cancer to allow biopsies to be accurately directed towards suspicious areas.” Additionally, studies have demonstrated that even when MRIs are of such high quality, the radiologists who interpret them vary in experience. Some are more accurate than others. Here’s where Artificial Intelligence can help.

Artificial Intelligence boosts MRI accuracy

At the 2023 meeting of the Radiological Society of North America (RSNA), an Artificial Intelligence (AI) breakthrough was presented by UCL Associate Professor of Radiology Dr. Francesco Giganti. He and his team developed and tested AI software capable of identifying biopsy candidates based on their mpMRI scans. When a patient is suspected of PCa based on his PSA blood test, what MRI factor determines if he should proceed to biopsy? It’s the image-based detection of clinically significant PCa, defined as Gleason score 3+4 or higher.

How can imaging calculate a Gleason score before biopsy? Well, mpMRI uses 3 parameters or imaging sequences to gain an anatomic and functional profile of a patient’s prostate gland:

• T2-weighted (T2W) sequences reveal the anatomy and structure of the gland, so any abnormality will show up.
• Diffusion-weighted imaging (DWI) sequences quantify tissue density in areas where the motion of water molecules is restricted, which characterizes PCa tumors.
• Dynamic contrast enhancement (DCE) reveals blood flow in the new blood vessels that tumors develop, creating inflammatory changes made visible by a contrast agent injected in a vein.

By using hundreds of specially marked MRIs from PCa patients to “teach” the software, it learns to read the factors that indicate not just the presence of a tumor but also its aggression level, and assign it a Gleason score (sum of two numbers). Dr. Giganti’s presentation noted that when the new software was designed and validated, it was tested on 794 patients included in the study. All patients had initial MRI followed by biopsy, so the team had access to the biopsy-based Gleason scores; 34% of the patients had biopsy-proven clinically significant disease. The team compared AI’s performance at identifying clinically significant PCa against the actual Gleason scores. As described in a news report, “The software automatically outputs scores intended to identify Gleason score ≥ 3+4 clinically significant prostate cancer.” As it turned out, the software specifically identified the correct patients with 94% accuracy, and was 95% correct in ruling out those who did not require biopsy. The team concluded that their proposed software performed at a level comparable to human radiologists who interpret mpMRI scans.

This is exciting news for both doctors and patients. Since demand has increased for mpMRI before biopsy, the workload for radiologists has likewise risen. AI software can take on a partnership role by flagging clinically significant disease. While the radiologist will always have the final say, AI’s preliminary reading will ease the stress of many scans to interpret in a day.

Thus, critical reports for patients with clinically significant disease are quickly relayed to their doctors in order to formulate treatment plans as soon as possible. And, by identifying those cases unlikely to carry significant PCa, AI expedites the time needed by the radiologist to review AI’s determination. The less strain and pressure on radiologic readers, the more attention they can focus on scans, and the more accurate their reading. In addition, AI can help even out the variations in performance between less experienced and more experienced readers.

Best of all, for patients in Robert’s situation, by ruling out significant PCa, AI can help countless men feel lucky when they can avoid an unnecessary biopsy.

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.