With an ever-increasing array of Artificial Intelligence (AI) applications in radiology, it’s a boom time for turning MRI into a powerhouse. In particular, both machine learning (ML) and deep learning (a particular type of ML) enable MRI of the prostate to branch out in new radiological directions. Out with the old, in with the new. When compared to conventional transrectal ultrasound, multiparametric MRI (mpMRI) “has demonstrated a better diagnostic accuracy and is becoming a clinical routine examination for patients at risk of having csPCa [clinically significant prostate cancer].”^[i]

Setting up ML programs for use in detecting and diagnosing csPCa requires “training” it to recognize MRI imaging features. It also requires the use of algorithms, which are basically sets of rules to be followed in calculations or other problem-solving operations. In a comprehensive review of ML applications in prostate MRI, Cuocolo, et al. define ML as a branch of data science:

based on the development and training of algorithms, by which computers may learn from data and perform predictions without previous specific programming. The main difference with classical rule-based algorithms is represented by their ability to take advantage of increased exposure to large and new data as well as to improve and learn over time.^[ii]

A subset of ML called deep learning (DL) has been created to mimic the ways in which the human brain learns and processes information, but DL requires larger datasets to train it. Both ML and DL programs are already trained and applied to increase the speed, efficiency, accuracy and consistency of radiologists’ work. Here are a few of tasks that ML and DL are capable of:

• Prostate segmentation, meaning identifying prostate zones in order to precisely locate suspicious lesions. Computer-automated segmentation has proven faster and at least as accurate as manual segmentation,
• Determining tumor aggression level to distinguish indolent cancers from clinically significant ones, which allows for treatment planning as well as qualifying patients for Active Surveillance and monitoring them for progression,
• Detecting/diagnosing tumors, including their zonal location, and applying the PI-RADS scoring system,
• Distinguishing between cancer vs. other prostate conditions such as BPH (benign prostatic hyperplasia) or prostatitis,
• Staging prostate cancer based on its extent, which aids in treatment planning, and
• Predicting biochemical recurrence.

One other interesting feature of ML/DL-based results is that many of the research studies that test the accuracy of AI findings were done without the use of a dynamic contrast enhanced (DCE) imaging sequence, yet still performed very well. MRI-DCE is done with a contrast image formulated with gadolinium. Prior to mpMRI that includes DCE, patient kidney function is evaluated to ensure that the kidneys are healthy enough to wash the mineral out of the body. In addition, DCE adds time and expense to the scan, so some urologists and radiologists are turning to biparametric MRI which uses only two imaging sequences. Does biparametric MRI miss key information about a patient’s prostate cancer? Studies have shown that with the addition of ML or DL, biparametric MRI demonstrates accuracy on a par with mpMRI. As Cuocolo, et al. reassuringly state, “ML could help in avoiding the systematic use of contrast agents for prostate imaging…”^[iii]

New ML and DL models continue to be developed, tested and marketed to prostate MRI centers. At the Sperling Prostate Center, we have incorporated AI programs for many years, making us early adopters and leaders in prostate imaging supported by AI. We happily look forward to even greater support for our excellence in clinical services as AI flourishes. You can find more blogs on AI in medicine at our companion website, Sperling Medical Group.

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.

References

[i] Cuocolo R, Cipullo MB, Stanzione A, Ugga L, Romeo V, Radice L, Brunetti A, Imbriaco M. Machine learning
applications in prostate cancer magnetic resonance imaging. Eur Radiol Exp. 2019 Aug 7;3(1):35.
[ii] Ibid.
[iii] Ibid.