Precision neuro-imaging will deliver personalised diagnostic and treatment approaches for patients with chronic neurological disease.
Artificial intelligence (AI) is expected to transform both research and clinical neuro-imaging through automation and increased productivity, enhanced reporting accuracy and the rapid identification of critical abnormalities.
At SNAC, we are committed to developing AI algorithms and software platforms that seamlessly integrate with clinical radiology workflows – providing value add for clinicians without a single additional keystroke or mouse-click. SNAC has equipped with extensive experience in deep learning technology, including classification, segmentation and augmentation, to solve problems in the domain of neuro-imaging.
The extent of focal brain pathology provides the clinician with crucial information about the status of a patient’s neurological disease. For example, the number and volume of ‘lesions’ in the brain of patients with multiple sclerosis is an important determinant of disease progression and response to therapy. However, manual segmentation of lesions by a trained radiologist is tedious and time-consuming. In clinical practice, therefore, usually only qualitative measures are provided in radiology reports.
A deep learning model with segmentation techniques can drastically reduce the time required to determine lesion volume – to just 3 seconds. This equates to a significant saving in time and cost, and makes it possible to use quantitative scan analytics in clinical practice.
This image shows a side-by-side comparison of MS lesion segmentation using manual methods (left) and a fully automated, AI-based lesion segmentation method (right) developed by SNAC.
Intracranial hemorrhage is a critical abnormality requiring urgent medical intervention. Images acquired in busy hospital or outpatient settings are usually reported by specialist radiologists in chronological order, which in some centres may lead to significant delays. Deep learning AI algorithms developed by SNAC, trained with a large dataset of ‘golden labelled’ scans annotated by certified radiologists, robustly detect acute intracranial hemorrhage in CT images within seconds. Automatic triage results are reflected on the clinician’s Radiology Information System (RIS), facilitating automated prioritisation of critical cases.
Volumetric analysis of specific tissue types may be confounded by the presence of pathology, such as white matter lesions in multiple sclerosis. These lesions have a heterogeneous appearance on T1-weighted MRI images that reflects the severity of tissue injury. Hypointense white matter lesions are often misclassified as grey matter or cerebrospinal fluid by automated segmentation algorithms, resulting in biased brain substructure volume assessment.
The use of AI-algorithm based ‘inpainting’ techniques to augment images prior to analysis can mitigate this issue, improving the accuracy and consistency of brain volume measurement results from automatic analysis pipelines.