Deep learning for neurological disease       

links

 Lilli-Marie Pavka photo

Deep learning for neurological disease - overview


Decoding mental states

Deep learning for neurological disease

IBM scientists use advanced machine learning techniques to develop new methods for the personalised treatment of epilepsy and Alzheimer’s disease patients.

Using brain-inspired chips and deep learning for epilepsy treatment

Epilepsy is the most common, chronic noncommunicable disorder of the brain that affects people of all ages. Approximately 50 million individuals worldwide suffer of this disease. Around 70% of these patients respond to treatment, whereas 30% remain in the dark with no warning signs or prevention methods of an epileptic seizure. This could mean, for example, that they are not allowed to drive or are limited in their career choices.

TrueNorth pipeline

IBM scientists are using machine learning algorithms and a neuromorphic chip called TrueNorth for the real-time analysis of electroencephalogram (EEG) data to develop new devices for epilepsy patients. The brain-inspired TrueNorth chip is powered by 1 million neurons and 256 million synapses made of silicon, yet it consumes less than 70 mW. This makes it ideal for applications in small and energy-efficient wearables with high compute capabilities. Our scientists envision an intelligent, wearable device which may identify signs of an upcoming seizure and alert a patient or automatically administer medication — seamlessly to the patient’s life. This could not only improve dramatically the lives of people with epilepsy, but also save lives.

Supporting the diagnosis and personalized prognosis for Alzheimer’s patients

The most common type of dementia is Alzheimer’s disease, which affects 70% of all dementia patients. The disease destroys brain cells and thus impairs the memory and thinking capabilities as well as the behaviour of a patient.

In the onset of the disease, symptoms can be very gradual, and the progression varies from patient to patient. Early detection is important to help rule out other illnesses with similar symptoms and allow symptomatic treatment to start as soon as possible. For the diagnosis, physicians use several examination methods, such as memory tests, medical images of the brain, and measurements of a biomarker in the spinal cerebral fluid. To support an early diagnosis, IBM scientists are identifying novel biomarkers using machine learning techniques to identify the key indicators of Alzheimer’s disease, such as memory loss, before clinical signs appear. This could make population screening of risks groups possible. Furthermore, scientists are endeavoring to understand the genomic and other modifiers of the cognitive decline associated with Alzheimer’s to support clinicians in making a personalized prognosis of the disease progression. To achieve this goal, IBM scientists are analysing the biochemical, neurological, genomic and brain imaging data of more than 800 participants of a long-term study.