Andrew Goldbaum 1/27/18 Current Event 14 C Even

“Artificial Intelligence Can Detect Alzheimer's in Brain Scans Six Years Before Diagnosis.” Neuroscience News, Neuroscience News, UCSF, 4 Jan. 2019, neurosciencenews.com/ai-alzheimers-detection-10427/.

Can artificial intelligence - a field specializing in teaching computers to learn from its experiences and use human-like cognition to solve problems - detect the progression of Alzheimer’s Disease years before the final diagnosis is made? New research has indicated that this is a very real possibility. For some context,  drugs have emerged in recent years to curb the progression of Alzheimer’s disease, but they are usually useless. This is because final diagnoses are made for Alzheimer’s after enough neurons have died to make the disease irreversible, and the drugs only work when Alzheimer’s is still early in its progression. To combat this problem, neurologists have attempted to interpret PET (Positron Emission Tomography) scans. These scans can isolate a specific molecule - say glucose - and display changes in their levels throughout the brain. Glucose levels are important to note for Alzheimer’s disease because the disease is characterized by dying neurons; and when neurons die, they stop using high levels of glucose to carry out life processes. Therefore, the fix is simple: catch glucose changes throughout the brain early enough in the Alzheimer’s progression for the drugs to still be effective, and death rates from the disease will plummet. However, this is the difference between “difficult” and “simple”: experts in the field already know how to interpret data to detect tiny changes in the point of interest of a disease - say the early development of a tiny tumor - but glucose changes are often so subtle and slow-progressing in Alzheimer’s that these experts may not see that mild glucose losses here and there are part of the overall early onset of a disease. However, by feeding vast quantities of PET scan data of patients who eventually developed either the disease, mild neurological impairment, or no symptoms to an artificial intelligence algorithm; it has learned to distinguish the subtle glucose changes that lead to Alzheimer’s from those that don’t with astounding accuracy: 92% accuracy in one set and 98% in the other, six years before the final Alzheimer’s diagnosis.

PET scans are already an extremely powerful technology: I intern for a psychiatric institute - The Amen Clinic - in New York City, and it uses this scan to detect blood flow levels to specific parts of the brain in order to determine if someone has ADHD, Cranial Cervical Syndrome (CCS) and epilepsy, autism, among other conditions. The psychiatrist I intern for already knows how to properly examine this data, diagnose the condition, and offer a highly specific treatment to alleviate much of said condition with high levels of accuracy. However, something I noticed is that the types of patients who come into this clinic tend to have the same few conditions: predominantly ADHD, CCS, and sometimes autism. If the diverse types of biological molecules PET scans can examine (from glucose levels to blood levels, etc.) can be augmented with the cognitive powers of deep learning AI algorithms, early onset diagnoses of a much wider range of diseases is within reach. Therefore, the entire arsenal of available treatments can be tailored for someone’s specific problem before it ever becomes that problematic.

This article was well-written because it clarified not just the new development in the field, but why exactly it was necessary: not only did the article discuss how AI can identify the slower, widespread changes that humans can’t identify; but it described everything from why glucose is used in the PET scans to the main gist of what the scans do to where the current treatments for Alzheimer’s are applied and their setbacks… The only explanation I found to be lacking, however, was that for the AI: the article should have stated whether it used supervised or unsupervised learning and if there are future applications of this kind of algorithm besides Alzheimer’s detection.