Researchers from University of California in San Francisco have developed an Artificial Intelligence tool that can predict Alzheimer’s disease years before the actual diagnosis. The new AI model could prove to be very useful in treating Alzheimer’s patients.
AI has come a long way from when it first became a thing. The first big breakthrough in AI was when a simple model could distinguish cats from youtube videos. It’s come a long way from there and to think that it can be applied to medical sciences as well is great. Deep learning was used and the AI was trained using positron-emission tomography (PET) images of 1,002 people’s brains.
If you’re not familiar with how AI training works, the data is divided into two categories. The first of the two is the training data, the other category is the test data. Usually, it’s a 90:10 split i.e. the training data is 90% of the total data we have whereas the test data is only 10%. This is done so that we avoid what is known as ‘overfitting’. Any AI model that overfits is considered a bad model as it won’t give us a realistic prediction to other data.
The images taken here were also fed to the algorithm in a 90:10 manner. The algorithm was taught how to spot different features of Alzheimer’s disease using these images and then it’s accuracy was tested using the test data. Once the algorithm was fully trained, they put it to a real test.
The AI was put to the test
PET images of the brains of 40 other people were taken and were fed to the AI for predictions. The algorithm predicted the individuals what would get a final diagnosis of Alzheimer’s and it was able to do so accurately. On average, it took 6 years after the scans for the predictions to come true.
In their paper, the team explained how their algorithm achieved 82% accuracy specifically at 100 sensitivity with an average of 75.8 months prior to final diagnosis. The model performed pretty well and anything above 80% accuracy is fairly good for a deep learning model.
Co-author of the paper, Dr. Jae Ho Sohn, who works in the university’s radiology and biomedical imaging department said:
“We were very pleased with the algorithm’s performance,”.
“It was able to predict every single case that advanced to Alzheimer’s disease,” he added.
Deep learning gets better with time
The great thing about deep learning and AI, in general, is that it gets better more data is available. It keeps learning which helps it get better in making its predictions over time. If, for example, it makes a wrong prediction, it uses the data it gets and learns from it in order to avoid the same mistake again.
Deep learning is all about learning through example, just as how humans usually learn. The algorithm basically teaches itself by looking at thousands of images that are labeled. The authors pointed out that:
“compared with radiology readers, the deep learning model performed better, with statistical significance, at recognizing patients who would go on to have a clinical diagnosis of [Alzheimer’s disease].”
Although the researchers have come up with something promising, they have warned that there is still a lot more validation that needs to be done since the study was done on a small scale. Further training of their AI model is required which will require much larger datasets i.e. more images contributed by more institutions.
However, Artificial Intelligence could indeed be a useful tool in the future which can help in starting the treatment process way before the final diagnosis. The end result can potentially save lots of lives. Moreover, the algorithm isn’t just limited to Alzheimer’s. The researchers plan on adding other types of recognition into the algorithm which would make it a highly versatile one.
To conclude, here is what study co-author Youngho Seo, a professor in the Department of Radiology and Biomedical Imaging has to say:
“If FDG PET with [artificial intelligence] can predict Alzheimer’s disease this early, beta-amyloid plaque and tau protein PET imaging can possibly add another dimension of important predictive power.”