Medical device developer EMVision Medical Devices Limited (ASX: EMV) has reported “very encouraging” findings from a new pilot clinical trial of its novel imaging technology.
The single-site study, at the Princess Alexandra Hospital (PAH) in Brisbane, of patients with diagnosed ischaemic or haemorrhagic stroke, is the first clinical study for the EMVision technology.
The primary endpoint was the collection of a dataset of stroke patients which improves the understanding of stroke on electromagnetic scattering effects in the brain. This end point has been met, producing datasets that have enabled EMVision to advance its imaging algorithm development and observe the correlation of EMVision scans with “ground truth” CT and/or MRI scans. Furthermore, clinician and patient feedback on the usability and comfort of EMVision’s clinical prototype has been collected.
The study was designed to collect data to tune the EMVision algorithms. The de-identified patient ground truth CT/MRI “training sets” made it possible for the algorithm team to refine the imaging and classification algorithms.
Additionally, clinical advisors in conjunction with the technical team identified five datasets where pathologies were estimated to fall outside of the anticipated prototype hardware range.
The EMVision classification was observed to demonstrate an ability to differentiate between haemorrhagic and ischaemic stroke with an overall accuracy of 93.3% [95% CI2] in the full sample (30) and 96% (95% CI) in the sample excluding patients with pathologies located outside the estimated prototype hardware range.
Localisation has been evaluated based on whether the EMVision fusion images resulted in target detection in the same quadrant as the ground truth scans (CT/MRI). For any scenarios where the ground truth image or fusion image had multiple areas of pathology identified, the clinical verifier has taken the most prominent / intense area to be the area of interest.
The EMVision fusion images were observed to be able to localise in the correct quadrant with an overall accuracy of 86.7% [95%CI] (full sample of 30 datasets) and 96% [95%CI] (sample with the 5 excluded datasets).
The study datasets have enabled the algorithm team to advance the hybrid “fusion” methodology, which is a powerful approach to imaging. The fusion hybrid works by extracting the target lesion and estimated location in each algorithm and applies a pixel-wise voting algorithm. The fused image then leverages the classification algorithm and can be overlaid on a predicted head template. The algorithm team will continue to advance this fusion methodology in consultation with EMVision’s clinical advisors.
Stroke neurologist and EMVision clinical advisor, Professor Michael O’Sullivan, said the latest results are exciting in showing excellent discrimination between ischaemic stroke and haemorrhage in the data collected to date.
“Interesting future questions include the sensitivity of the technique for early detection of bleeding, both in the pre-hospital setting and in stroke units, where monitoring is currently limited to detection of clinical deterioration.
“The value of combining multiple algorithms in a fusion approach is also an emerging theme, which could help to tailor the approach to the setting and clinical question.”
EMVision’s CEO, Dr Ron Weinberger, said the promising results, the first in stroke patients for the company’s technology, provide a strong foundation with which to progress its development program.
“We are delighted by these results which indicate that we are able to discriminate and localize haemorrhagic and ischaemic strokes with an encouraging degree of accuracy under these conditions.
“We have exceeded our original objectives for this study. While we still have a way to go, we are well placed to develop our value proposition into a fully-fledged commercial product.”
