Scroll Top

How will AI/ML Driven Image Analytics change the future of Neuroimaging?

Consider the typical way someone with a disorder, such as phenylketonuria, diabetes, or asthma might be diagnosed: the patient exhibits some number of symptoms and after an exam, the provider might order a series of blood, genetic, or lung function tests. One thing these tests likely have in common? The results are unambiguous. A genetic variant or enzyme is either present or not present. The ratio of white to red blood cells or blood glucose level can be measured. But for patients who present with a potentially broken bone, breast cancer, or neurological disorder, such as multiple sclerosis, the diagnostic pathway may be more complicated due to the subjectivity involved in reading medical images. Is the barely perceptible difference in one area of a mammography image breast cancer, something benign, or an artifact of the machine and image processing? Does the brain scan show a structural defect that might explain the patient’s symptoms?

Admittedly, for the past several years, radiology has made tremendous strides in addressing the subjectivity issue through digitization. By digitizing the scans, radiologists can more easily compare images taken at different timepoints, such as before and after treatment, or yearly to assess whether a condition has improved, stayed the same, or gotten worse. Perhaps even more importantly, this digitization allows for quantificationof these kinds of differences—making interpreting radiological images more like measuring an analyte in a blood sample. Newly emerging are software and products based on artificial intelligence (AI) that can analyze radiological images more rapidly and often, as well as or even better than, the doctor. For example, earlier this year, FDA approved an AI-based device that can diagnose diabetic retinopathy by non-specialists and another to be used as an adjunct diagnostic to help providers better diagnose wrist fractures in adults [1]–both of which are tools that use images captured by devices for assessment.

In my book, MoneyBall Medicine: Thriving in the New Data-Driven Healthcare Market, I interviewed Christoph Wald, Chairman of the Department of Radiology at Lahey Hospital & Medical Center in Boston who described his department’s fully digital transition [2].  The consequence is a department that has a defect rate (a deviation from the expected image quality that interferes with image interpretation) of less than 1% and with many newly automated processes that drive improved efficiencies. But getting to the improved image quality and interpretation in the first place is a challenging task.

On Episode 9 of the MoneyBall Medicine Podcast, I talk with Wim Van Hecke, CEO of Belgium-based Icometrix. Launched in 2011, Van Hecke approached radiology from an engineering perspective, developing software solutions to mine as much quantifiable data as possible from medical imaging. The result is a suite of products for a variety of disorders: multiple sclerosis, traumatic brain injury, and dementia, with the goal of providing physicians with usable data to ascertain a patient’s brain changes over time and corresponding treatment effectiveness. Van Hecke and I discuss why Icometrix focuses on brain disorders and how it uses technology, like cloud computing, to scale. We touch on the role of regulation for AI-based imaging technologies and the importance of flexibility to adjust to market needs. Join me for this episode of the MoneyBall Medicine Podcast!


  1. Tuma, R. S. 2018. FDA Approves AI-Based Software for Wrist Fracture Detection. Medscape.
  2. Glorikian, Harry; Branca, Malorye Allison. Moneyball Medicine | Thriving in the New Data-Driven Healthcare Market. 1 ed: Productivity Press, 2017. Print.