AI is set to bring fundamental change in medicine and healthcare practices. This ranges from studying diagnostic patient data such as ECG, X-ray, and EEG with the help of machine learning to detect diseases at an early stage based on slight changes.
However, implanting of AI-based implants in the human body is still a key technical challenge. Scientists at the Chair of Optoelectronics, TU Dresden have for the first time succeeded to develop a bio-compatible implant. Elaborately, the AI platform classifies healthy and pathological patterns of biological signs such as heartbeats in real time. In addition, the platform detects pathological fluctuations even without medical supervision.
The results of the finding is published in Science Advances.
The initiative demonstrates an approach for classification of diseased and healthy bio-signals based on a biocompatible AI chip in real time. This involved the use of polymer-based fiber networks that resembles the human brain structurally and enables neuromorphic AI principle of reserved computing.
In fact, the random array of polymer fibers forms a recurrent network. This allows the brain to process data that is analogous to the human brain. Besides this, the nonlinear layout of the networks enables to amplify even the minutest signal changes. For example, smallest change in the heartbeat are often difficult for doctors to evaluate. However, the nonlinear transformation using polymer network makes it possible without any obstacles.
Meanwhile, clinical trials to investigate the AI platform differentiated healthy heartbeats from three common arrhythmias with 88% accuracy rate. During the process, the polymer network used less energy than used by a pacemaker. The potential applications for AI systems that can be implanted are manifold.