The first task of this POC grant was to validate our device capable of millisecond precision synchronization of visual stimuli, patient button presses and neurophysiology (e.g. EEG, electromyogram (EMG)) signals (coined SyncR), as well as to validate our SSRT protocol. We ensured full functionality and failsafe operation and manufactured five functional prototypes. While SyncR is protected as a utility model, we sought to broaden IP protection by submitting an international PCT application, according to our IPR plan. Our PCT application has reached the national phase, in which we seek to protect the device in Europe, US and China, covering the largest potential markets. We carried out full market research including SWOT/PESTLE analysis, market assessment, strategy planning, market entry strategy, manufacturing strategy, business planning and cost structure analysis, branding and distribution.
Next, we performed SSRT testing after recruiting n = 15 adult, non-parkinsonian volunteers to improve the task paradigm we developed. The following main questions were posed regarding the suitability of the paradigm. (i) What is the correct strategy for stop signal delay (SSD) adjustments during the task? (ii) What is the optimal proportion of stop trials? (iii) What is optimal trial number to achieve accurate SSRT estimates? Task parameters were determined and optimized based on the behavioral results.
In agreement with our project plan, we performed real-world testing of QCT and SyncR on 15 patients with Parkinson’s Disease, recruited by our neurologist partners at the Neurology Clinic of Semmelweis University. Patients were tested in 10-minute sessions first using their right hand, then using their left hand. The SSRT task was combined with 60-channel EEG recording and EMG recording of lower arm muscles that control finger movements. We found that the SSRT task was capable of quantifying cognitive parameters including inhibitory control, impulsivity and motor learning. The SyncR device was capable of millisecond precision synchronization of visual stimuli, electrophysiology signals and patient reactions in a real-world scenario.
In addition to standard spectral based EEG analyses, we performed machine learning analysis of EEG recordings from patients with Parkinson’s Disease in three phases. In the first phases, a literature and database search were performed to assess state-of-art AI models and publicly available datasets. Datasets were filtered based on relevance in consultation with a neurology expert. In the second phase, a foundation model for EEG was constructed. Recordings showed clustering by disease labels but not by recording length or by source datasets. These confirm that project-goal-relevant features were encoded in the recording embedding, thus validating our foundation model. In the third phase, we included novel EEG data recorded in course of this POC project. The classifier trained on the public datasets achieved over 90% performance, confirming that EEG collected during the SSRT task carried information on neocortical features relevant to Parkinson’s Disease.
