Artificial Intelligence Predicts Alzheimer’s In Brain Scans Before Diagnosis

Artificial intelligence can soon become a blessing for people who have Alzheimer’s or about to develop the ailment sometime later in life. Although the disease can’t be cured, if Alzheimer’s is detected in early stages, its progress can be halted due to advancement in medicine. However, detecting the disease in early stages is the bottleneck stopping radiologists.

If artificial intelligence can help scientists detect the illness sooner than later in coming years, it can give them an opportunity to intervene and stop the progress.

The Problem

The Resident Physician, Jae Ho Sohn, MD, MS, who found out the solution says that the biggest hurdle in the treatment of Alzheimer’s is that by the time it is detected, the damage is irreversible. The clinical symptoms and accurate diagnosis that helps Doctors predict the presence of Alzheimer’s are carried out at a stage when considerably high neurons have already died.

The Research

In a study carried out by Sohn, he tried to predict if an individual will encounter Alzheimer’s disease after they experience memory impairment for the first time. This time is also the best possible time for intervention with probable treatment and medication.

To detect the presence of Alzheimer’s, most common diagnosis is done through PET (Positron emission tomography). These PET scans measure the presence of specific molecule levels in the brain. For instance, glucose, which helps in diagnosing the ailment as it is the main fuel source for our brain cells. An active cell uses more and more glucose for effective functioning. But, diseased cells start using less glucose and eventually stop using any when they die.

In the study, scientists used this information and PET scans to evaluate the difference in glucose levels across the brain. Another major fact that plays a role in the diagnosis of Alzheimer’s is that this ailment is a slowly progressive disorder. This means that the changes are slow and slight and not visible to the naked eye in the scans. Hence, even when the glucose levels in the brain keep changing slowly, it is not diagnosed as early as we should predict it.

So, to diagnose Alzheimer’s early, radiologist used artificial intelligence algorithm to figure out changes in frontal and parietal lobes. They took starting data from Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. These PET scans were diagnosed with no disease, mild cognitive impairment, or Alzheimer’s disease. When this data (1921 scans) was given to the algorithm, it started predicting how the disease progresses and what are the slow changes observed in the brain.

After this, the radiologists fed two datasets to the algorithm: one from the ADNI only but these scans were not previously given to the algorithm and other from UCSF Memory and Aging Centre’s data.

In the first test, the algorithm predicted the presence of Alzheimer’s correctly in 92% patients and the second time, it accurately detected 98% patients. In fact, an amazing thing was that this algorithm predicted the presence of Alzheimer’s 6 years before the actual diagnosis. Starting the treatment 6 years earlier – that’s some progress radiologists were looking forward to.

Conclusion

Although Sohn still believes that we have a long way to go because the datasets were small, the progress is extremely useful. These radiologists will try to recover more information in this respect in 2019, but it amazing how we have made the progress. Sohn further noted in the experiment that radiologists are great at predicting small focal changes such as in case of the brain tumor. But, finding changes in the global region is a struggle. The artificial intelligence is known to carry out such tasks, hence, can immensely improve the diagnosis stage and treatment stage of Alzheimer’s disease.