If your doctor and an algorithm arrived at two different diagnoses, which would you trust? Of course, it depends on the specific context but this question opens a much needed discussion about a transformative process occurring in medicine: computers are beginning to perform tasks of physicians. While modern medicine utilizes medical technology primarily as an aid for physicians, future technology may afford diagnostic capability that rivals that of humans. The question is, in what capacity do we incorporate such advances into our clinical practices?
Modern medicine has vastly expanded due to advancement in medical technology: imaging equipment allows doctors to see inside the body at a remarkably fine resolution, surgical instruments allow certain complex procedures to be performed with no post-operative marks on the patient other than a pinhole, implanted devices such as pacemakers offer intervention that would otherwise be impossible, and healthcare IT software manages patient data across a clinical workforce rapidly and efficiently. There’s no arguing that technology has increasingly become an integral part of the clinic. But what if this technology started making decisions?
Decision making, in computer terms, is determined by the algorithm – a process that integrates inputs in order to produce an output according to a defined set of rules. This is similar to a “clinical algorithm” used by physicians to make sense of patient data. The rapid advancement of computation has given rise to more elegant and more complex algorithms that harness the computer’s inherently superior ability to make rapid calculations and handle large sets of data.
Using computers for automated diagnosis may sound like something out of a sci-fi movie featuring robot doctors that diagnose and treat patients. We are far from that idea; however, certain specialties of medicine, radiology and pathology amongst them, are amenable to automated diagnostic technology. It takes a carefully trained eye to spot certain signatures of structural or functional pathology in patient images. Biomedical image analysis is a developing field of research that aims to utilize automated computerized processes to spot those same signatures. The computer “looks” for certain image features just as a radiologist would, however it can do so with greater speed and objectivity. Further, it can provide quantitative information which a human is not able to do. When these algorithms develop into more clinically robust tools, they will be able to define certain disease markers as they appear in various imaging studies. Whether or not the algorithm’s output is the definitive diagnosis remains a point of contention, so for now, this type of technology serves to guide physicians.
Fundamentally, algorithms are mathematically perfect however, technology is not. Error due to malfunction of any part of a given system is potentially disastrous especially when dealing with someone’s health. I liken the situation of automated diagnostic technology to that of self-driving cars. At present, we have prototypes that work, and a concept that could potentially eliminate all traffic problems (if all decisions were automated and in-sync, there may be no accidents). The passengers in the car entrust their safety to a machine, so system failure is a sensitive matter. Safety mechanisms of course are part of the engineering design process and should also be a part of the clinical diagnostic process with any kind of future technology. This is why at present, automated diagnostic technology serves only as data for the physician to review.
Artificial intelligence, neural networks and machine learning are some of the hot topics being refined in the world of computer science for application to a myriad of different problems including those in healthcare. There are benefits of using automated diagnostic technology in the clinic but at present, such technology is not advanced enough for isolated use and therefore acts simply as an aid to the physician. As proof of concept solutions emerge from industry and academia, we must be prepared to discuss their use as isolated instruments in the clinic .