Artificial Intelligence in Medicine: Taking MRI Interpretation to a New Level

When a person is born, is their IQ fixed for life? IQ stands for intelligence quotient, and is a standard measure of mental performance. Is it fixed, or can it be taken to a new level as the person develops? This question has sparked all manner of research with infants, young children, school-age kids, etc. It appears that at least until adulthood, a person’s built-in mental capacity can indeed grow somewhat.

What all neurological sciences agree on is the importance of education. Some brains may inherently be smarter than others (e.g. Einstein) but a healthy brain can always learn more than what it knows. You would not expect to teach algebra to a second grader, but their cumulative learning prepares them for more abstract and sophisticated logic and formulas in the eighth or ninth grade when algebraic thinking is introduced. Step by step, the more content they learn, the more they can compute.

In a parallel way, Artificial Intelligence (AI) becomes smarter as it is trained on increasing amounts and complexity of content. When applied to reading MRI scans of the prostate, AI programs have mastered identifying suspicious areas of images the way that an elementary school youngster has mastered the arithmetic processes of addition, subtraction, multiplication and addition. Those skills form a foundation for more complex formulas.

Thus, AI’s current state of accurate image-based identification of suspected tumors is the foundation for more sophisticated calculations. Now, a new study illustrates a more complex process that moves toward a new level of actual PCa diagnosis. It does so by adding in clinical factors (patient age, PSA, prostate dimensions and volume, biopsy Gleason score, and PI-RADS from previous MRI) to its image reading.

The study team designed a Deep Learning program that can handle a large number of different types of data applied to each case. They write, “Deep learning is designed based on the function of the human brain and can extract features alone and find non-linear relationships that cannot be understood by humans.”^[i] Their process of classifying patients according to a 5-point scale (1=benign, 5=cancer) relied on four imaging sequences, or parameters, of each patient’s multiparametric MRI (mpMRI) scan as follows:

… a quadruple sequence of mpMRI images is analyzed and then clinical and pathological data are added to them to determine the level of cancer from benign to malignant with grades 1 to 5. In the initial stage, data has been collected and pre-processed. In the next step, four separate neural networks are trained for image analysis. The layers containing the features extracted from the images along with the clinical and pathological data are given as input for the final neural network (a fully connected network).

Their conclusion sums up the comparison between an AI model that did not involve clinical factors vs. their integrated Deep Learning diagnostic model. “In the model without clinical and pathological data, the maximum accuracy reached 88%, but by adding these data, the accuracy increased to 96%, which shows the significant impact of clinical and pathological data on the accuracy of diagnosis.” In short, we applaud them for taking AI interpretation of prostate MRI to a promising new level.

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] Sherafatmandjoo, H., Safaei, A.A., Ghaderi, F. et al. Prostate cancer diagnosis based on multi-parametric MRI, clinical and pathological factors using deep learning. Sci Rep 14, 14951 (2024).