Taiwanese firm HippoScreen is using Intel’s OneAPI tools and AI frameworks to optimise deep-learning models that help diagnose depression.

Using the Intel AI analytics toolkit and OneAPI base toolkit, HippoScreen improved efficiency and build times of deep-learning models used in its brainwave artificial intelligence (AI) system. These improvements enabled it to broaden applications to a wider range of psychiatric conditions and diseases.

Globally, an estimated five per cent of adults suffer from depression. There is no one-size-fits-all diagnostic procedure for depression and, while some cases can be clinically diagnosed, most assessments are dependent on patients’ subjective self-descriptions.

To overcome this problem, along with the widespread stigma surrounding depression, HippoScreen developed the SEA stress EEG assessment system, which helps doctors more accurately diagnose mental health conditions. The system includes an electroencephalogram (EEG) amplifier for data collection and signal processing, a graphic user interface for test process control, and an AI algorithm for data analysis.

SEA analyses 90-second brainwave signals and provides an objective and quantifiable evaluation index that aims to represent numerically the probability that an individual is suffering from depression.

To improve algorithm efficiency and diagnostic accuracy while reducing the delivery times of critical diagnostic results to medical personnel, HippoScreen leveraged the optimisations of Intel analysis tools and AI frameworks. Using the Intel VTune Profiler analysis tool, the company reached maximum performance and minimum CPU use with a thread count of five.

In addition, performance doubled, allowing the company to identify and resolve threading oversubscription issues quickly.

Intel Optimization for PyTorch and Extension for Scikit-learn, alongside HippoScreen’s proprietary algorithms, analysed system EEG data features that culminated in a unique decision factor and resulted in 2.4 times performance improvements.

“We at HippoScreen have been able to take advantage of the software optimisations in Intel Extension for Scikit-learn and Optimization for PyTorch to accelerate the build times for the AI models in our customised EEG brain waves analysis system by 2.4 times,” said Daniel Weng, chief technology officer at HippoScreen NeuroTech.