This artificial intelligence system was able to detect 62% of actual cases of cognitive decline and memory impairment in initial practical tests, according to the Massachusetts General Hospital Network's public relations department.
Assistant Professor Hossein Estiri of Massachusetts General Hospital said: "We have not just developed an artificial intelligence system, but a complete clinical set of neural networks that includes five specialized agents, each of which critically examines the results of the other components of the system and helps them reach more accurate and complete conclusions, like what happens in a consultation meeting of physicians."
The doctors said: "Dementia is usually discovered in patients in the middle or late stages of development, when the pathological processes in brain tissue are so complex that they are difficult to suppress using various treatments or lifestyle changes. This is partly because its diagnosis requires various arduous and lengthy cognitive tests, which are extremely difficult to conduct on a national population scale."
Clinicians hypothesized that the progression of dementia might significantly impact other clinical indicators that physicians record during routine patient examinations or in other contexts, and which are maintained in individual patient records. To search for and analyze these indicators, researchers developed a multi-component artificial intelligence system based on two different versions of the LLaMA large language model and the Med42 medical neural network.
To train it, the researchers compiled a collection of several thousand physician observations and recordings of clinical procedures performed while monitoring the health of two hundred people with dementia and healthy individuals. This information was also reviewed and analyzed by several leading experts in dementia diagnosis, whose advice and diagnoses were used in training and testing the AI system.
Subsequent testing of this system on another set of patient data showed that it was able to detect approximately 62% of cases of early-stage dementia, with the AI system correctly identifying a positive diagnosis in 98% of cases. The AI system also detected existing dementia in 58% of cases where leading experts had incorrectly made a negative diagnosis, significantly expanding the practical applications of such intelligent systems.
