The advance and prevalence of artificial intelligence may largely assist disease diagnosis. Latest researches found AI can help diagnose eye diseases and Alzheimer’s in early stage and with higher efficiency and accuracy.
Artificial intelligence technology can help diagnose eye diseases more efficiently
According to Australian media reports, an international research team led by Australian researchers has developed a new technology that uses artificial intelligence to analyze retinal images generated by ordinary optometry equipment to efficiently identify diabetic retinopathy with an accuracy rate of 98%.
Diabetic retinopathy is one of the complications of diabetes and a common blinding eye disease. The cause is that the high glucose toxicity caused by diabetes brings changes in the ocular nervous system and capillaries, and the diagnosis of the disease at an early stage is essential for protecting the patient's vision.
A team of researchers from Australia and Brazil developed an image processing algorithm based on artificial intelligence technology. By analyzing the retinal images taken by the fundus camera, the key features of diabetic retinopathy can be automatically and instantaneously identified, that is, the leaking liquid in the eyeball caused by the fact that retinal capillaries are broken. This is 98% accurate.
Kant Kumar, a professor at the Royal Melbourne Institute of Technology in Australia, who participated in the study, said that the commonly used methods for detecting diabetic retinopathy are expensive and invasive, such as the need for professional "eye optical coherence tomography". Relatively speaking, the newly developed test method is fast and economical, and is more suitable for promotion in underdeveloped areas where medical conditions are lacking.
According to reports, the researchers will discuss cooperation with fundus camera manufacturers to further improve the technology. Relevant research results are published in the latest issue of the British Journal of Computer Applications in Biology and Medicine.
AI power: six years ahead of the diagnosis of Alzheimer's disease!
Early diagnosis has always been a major difficulty in the field of Alzheimer's disease. Recently, researchers from the University of California, San Francisco, using machine learning algorithms developed by commonly used brain scanning technology, can diagnose Alzheimer's disease 6 years earlier, which brings new hope for early disease intervention.
Alzheimer's disease is a degenerative disease of the nervous system. Patients with Alzheimer's disease may experience symptoms such as memory impairment and aphasia, and their condition will gradually become worse and cannot be reversed. Although there are no methods to cure Alzheimer's disease, several new drugs have been used to delay the deterioration of the disease in recent years. The sooner you can intervene in a patient, the better the treatment will be.
The current difficulty in diagnosing Alzheimer's disease is that, with existing diagnostic criteria, a large number of irreversible deaths have occurred in the patient's nerve cells at the time of diagnosis. Treatment started at this time, and the results were not satisfactory.
Fortunately, a study led by Dr. Jae Ho Sohn, a radiology resident at the University of California, San Francisco, combined machine learning algorithms with positron tomography (PET) to make early diagnosis of Alzheimer's disease, which can be about 6 years and 4 months ahead. The study was recently published in the book Radiology.
Orthogonal tomography (PET) technology can detect the content of specific molecules in the brain, such as glucose. As a substance that supplies energy to cells, the amount of glucose in the brain can indicate the degree of activity of brain cells. For patients with early Alzheimer's disease, as the brain cells gradually die, the glucose content in the corresponding area will gradually decrease until it disappears completely.
Using PET scans to measure the decline in glucose levels in the brain (especially the frontal and parietal), Alzheimer's disease can be found or detected earlier. However, because the differences in PET images are very subtle, the naked eye cannot distinguish and make judgments. At this point, deep learning can make full use of its strengths, and take the role of doctors to analyze the image results of PET scans.
The researchers used public data from PET imaging in the Alzheimer's Neuroimaging Program (ADNI) as machine learning data, including PET scan images of patients with confirmed Alzheimer's disease or mild cognitive impairment.
After 1921 images of training, the machine learning algorithm has been able to accurately predict the presence of Alzheimer's disease through PET scan images. In the test, the researchers used an additional 188 images from ADNI and 40 images from the Center for Memory and Aging at the University of California, San Francisco. The diagnostic accuracy of machine learning algorithms is as high as 92% and 98%, respectively. Even more surprising is that the diagnosis of machine learning algorithms is 75.8 months earlier (equivalent to about 6 years and 4 months) than the existing diagnostic methods.
PET scanning is commonly used in clinical practice in combination with CT scanning and is inexpensive. Since PET scanning equipment is commonly used in primary medical institutions, this diagnostic method is more widely promoted. Starting treatment at a relatively early stage can also better inhibit the deterioration of the condition and benefit the patient.
In the field of brain nerves, deep learning has many functions that radiologists are difficult to achieve, especially in discerning global and subtle image changes. Dr. Sohn, the lead in the study, said that the machine learning algorithm will be further validated and calibrated to have the ability to diagnose more patients.
This article is compiled by scientists at BOC Sciences. The last decade has seen greater engagement of AI in medical field. This technology can be waived into many aspects of drug development from drug discovery to the screening libraries of lead compounds to drug design, DNA encoded library, virtual screening or even chemical synthesis like carbohydrate synthesis, fluorescent labeling, bioconjugation, etc.