Imaging Informatics: How AI Is Tackling Alzheimer’s Disease
By Dongmin Kim, PhD
Vol. 20 No. 11 P. 6
According to the World Health Organization (WHO), 50 million people around the world have dementia, and the condition strikes 10 million more each year. Alzheimer’s disease, the leading form of dementia, accounts for 60% to 70% of all cases.
Alzheimer’s is the sixth-leading cause of death in the United States, and it’s estimated that roughly 10% of the general population suffers from it. By 2050, the population of Americans older than age 60 will double. This demographic change will bring a drastic rise in the number of people living with Alzheimer’s disease, which will increase the burden on the health care system.
Since there’s no medical cure or treatment to reverse the progression of the illness, it’s paramount that health care institutions focus on early detection and subsequent treatment, which has been shown to slow the onset of dementia. Research on the first detectable signs demonstrates that the biomarkers predicting the disease may be spotted as long as 25 years before characteristic symptoms first appear. The challenge in providing early detection is finding cost-effective, noninvasive methods that can be applied to the general health care system.
Traditionally, physicians relied on medical history, tests, exams, and personal information to make a potential diagnosis of Alzheimer’s disease. Symptoms such as memory impairment, confusion, difficulty with communication, and poor reasoning are clear signs that neurons have already been damaged. However, by the time the obvious symptoms emerge, it’s too late to apply effective treatment because medication cannot reverse the damage to atrophied brain cells.
Early detection is a challenge, but a number of studies are underway, giving rise to promising innovations. Advancements in AI are of particular interest because they offer the potential for timely treatment and early intervention based on a combination of machine learning algorithms and neuroimaging technology. Through AI, millions of people could benefit from much-needed intervention protocols that can slow the destructive progression of the disease.
AI is utilizing the accelerated power of machine learning to sort population data, allowing doctors to make predictions that were never before possible. Large data sets created from millions of electronic records, health care claims, and genetic information, layered with geographic, socioeconomic, and lifestyle factors, streamline the identification of connections between data sets that may seem unrelated.
One of the main challenges with this kind of data assimilation is the variety of formats that can’t be easily merged, making it difficult to produce meaningful results. Data integrity is not only crucial in establishing credibility and accuracy but also an essential base for analytic systems.
In the Alzheimer’s field of study, some of these challenges have been overcome. Positive results have been obtained in algorithms that have compared specific biomarkers of the disease. A number of research initiatives have also shown positive results utilizing biomarkers and MRI, as well as genetics and clinical data merged into algorithms, with some displaying astounding accuracy with high-performing results. Specifically, since differences in glucose uptake are predictive of Alzheimer’s disease, creating a predictive algorithm that can spot glucose anomalies by merging PET scans with machine learning makes early prediction a possibility.
Detection With PET Scans
Since human technicians are unable to identify the incredibly subtle patterns hidden within complex medical imaging data, AI offers an advantage in this area. When it comes to predictive diagnosis, machine learning can employ medical imaging and data analytics to craft a diagnostic framework physicians can use to implement evidence-based solutions.
In Alzheimer’s patients, a decreasing amount of glucose in brain cells is apparent when the cells become diseased and die. As glucose levels drop, PET scans are able to map this slow progression. In particular, 18-FDG PET scans, which are also used to identify some types of cancers, can read these low levels of glucose. With this information and the application of machine-learning algorithms to PET scans, researchers can measure these molecules.
Applying the Algorithm
To efficiently diagnose early-stage Alzheimer’s disease, predictive algorithms must be trained. To accomplish this, researchers set up clear instructions on what the algorithms must look for and then upload an extensive public data set of PET scans from patients who were either in the normal control group or diagnosed with the disease or other cognitive impairments. As the algorithm runs through the process, it establishes the learning criteria needed to predict which of the groups will develop Alzheimer’s. After this training period, scientists apply further data sets from the past in order to evaluate the outcomes. This process has shown astonishing results, proving that deep learning algorithms can predict Alzheimer’s six years faster than traditional clinical diagnosis.
Since slowing early-stage Alzheimer’s is a race against the clock, these algorithms will give physicians a chance to treat the disease before irreversible, widespread brain atrophy has caused significant loss of brain volume. More studies are being developed to confirm the possibilities of diagnostics with AI biomarkers of Alzheimer’s, such as the abnormal buildup of proteins. The promising developments will compound the predictive power of AI.
Positive Net Benefits
In addition to improving outcomes for Alzheimer’s patients by making impairment more manageable, AI presents other significant advantages such as social benefits and net savings. For family members, providing long term care for patients is an expensive and exhausting undertaking that quickly depletes financial resources and becomes an added financial burden on the state. Early detection and treatment will not only help minimize the emotional strain caused by the disease but also offer notable cost savings for patients’ families and the health care system. A study by the Rand Corporation shows that “the total per-person cost of dementia care in the United States is about $42,000–$69,000/year. Most of this cost is related to caring for patients in the more severe stages of dementia who require institutional and home-based assistance with activities of daily living.”
By keeping patients at less severe stages of the disease for a considerably longer duration and using early therapy treatments such as cholinesterase inhibitors and other innovations, care costs and hospitalization will be greatly minimized.
AI can provide an early diagnostic tool with significant benefits that can improve outcomes for patients and their families as well as drastically lower the costs of health care. Applying the predictive power of algorithms is a crucial method that will enable doctors to provide proactive management and prepare for the inevitable growth of the baby boomer population while opening the door to massive savings for the US health care system.
— Dongmin Kim, PhD, is the CTO/director of the AI R&D Center for JLK Inspection, a Seoul, South Korea–based medical solutions provider specializing in AI-based technology. JLK Inspection’s universal AI platform is created by a combination of big data, experts and its own unique engines and algorithms, providing on-site/real-time service, seamlessly connected to all systems.