EMVision Medical Devices (ASX:EMV) has achieved "exceptional" results for its world first neurodiagnostic technology to identify stroke and stroke type.
In total 307 participants were enrolled in the EMView trial and successfully scanned with the EMVision emu brain scanner across Liverpool Hospital, Royal Melbourne Hospital and the Princess Alexandra Hospital. This included 277 suspected stroke patients with 48 haemorrhages and 140 ischemic confirmed strokes.
The study’s primary objective of collecting matched EMVision emu and CT/MRI ground-truth brain scans, for the advancement of the device’s AI-based diagnostic ‘blood or not’ and ‘clot or not’ algorithms, was achieved. The results of the ‘EMView’ multi-site study with the point-of-care emu brain scanner will be submitted for publication following peer review.
To train the AI algorithms, over 240 participants’ scan data was used (including confirmed haemorrhagic and ischaemic strokes, a variety of common stroke mimics along with healthy volunteer scans). This enabled the algorithm to learn patterns and features associated with a normal versus abnormal brain, including hyperacute, acute, subacute and chronic ischaemic and haemorrhagic stroke. The training set was diverse, with varied cases (across demographics, stroke sizes and locations, and time from onset) to enhance the model's generalisability. The neurodiagnostic AI algorithms were then applied to a test dataset that was isolated and not used in the training. This prevented the model from "learning" any patterns from the test data, ensuring the assessment provided a real-world evaluation with truly unseen cases, albeit on a smaller sample size, prior to definitive sensitivity/specificity confirmation in the upcoming validation (pivotal) trial.
The learning curves presented above illustrate the AI model’s improving recognition of a wide range of dielectric patterns and anomalies with additional clinical training data, resulting in steady sensitivity (true positive rate) and specificity (true negative rate) improvements. Meaningful performance gains are observed as more data is added, indicating the AI model’s scalability to continue improvement with expanded data. Following this exercise, the isolated test data is then added into the complete training dataset.
Next steps
The promising results of the EMView study allow EMVision to confidently proceed with the validation trial, as planned, where diagnostic performance of the portable brain scanner will be definitively demonstrated to support an FDA De Novo clearance.
EMVision has site visits scheduled this month with its prospective US investigational sites. The total direct costs of the validation trial, to be funded from cash reserves, are anticipated to be in the region of $4 million, subject to finalising contracts. The validation trial duration is estimated between six and 12 months.
"The results are very encouraging, particularly as related to detection capabilities and sensitivity to small haemorrhages. We look forward to confirmation of this impressive neurodiagnostic capability in the validation trial,” Co-chairs of the Australian Stroke Alliance, Professors Geoffrey Donnan and Stephen Davis, said.
