E-Bulletin, Issue 2 from May 5, 2020

European Union extends medical devices regulation compliance date; Free dataset with 874,035 brain hemorrhage CT images available for non-commercial use

Welcome to our introductory monthly newsletter on neuroradiology and AI. We hope you find it helpful. Please feel free to share it with others.

If you would like to be informed when the new issue is out, please sign up here.

Regulation change: Compliance date changed for medical devices Regulation 2017/745

The European Parliament and Council of the European Union passed Regulation 2020/561, which amends Regulation (EU) 2017/745 on medical devices. The new Regulation gives a one year extension for complying with 2017/745, due to the need for member countries to focus on the COVID-19 pandemic. The 2017/745 Regulation established a regulatory framework to “ensure the smooth functioning of the internal market as regards medical devices covered by that Regulation.” It sets high standards for medical device quality and safety, while reinforcing key elements of the existing regulatory approach of Council Directives 90/385/EEC ( 3 ) and 93/42/EEC ( 4 ). These include supervision of notified bodies, conformity of assessment procedures, clinical investigations and clinical evaluation, and vigilance and market surveillance, while introducing provisions to ensure transparency and traceability of the covered medical devices for health and safety improvement.

The new Regulation changed the enforcement date from 26 May 2020, to 26 May 2021.

 

AI Dataset CreationConstruction of a machine learning dataset through collaboration: The RSNA 2019 brain CT hemorrhage challenge

The Radiological Society of North America (RSNA) 2019 Brain CT Hemorrhage Challenge created the largest public dataset of brain CT studies to use in future machine learning (ML) applications, with 874,035 images. It was released under a noncommercial license, available to the ML research community to create high-quality algorithms for diagnosing intracranial hemorrhage. This paper, published in Radiology: Artificial Intelligence, shares how the dataset was collected and annotated.

The dataset includes a large variety of cerebral pathologic states (including subarachnoid, intraventricular, subdural, epidural, and intraparenchymal hemorrhage), with CT studies from multiple scanner manufacturers at three institutions, and in two countries. It includes scans taken from both emergency and in-patient settings at Stanford University in Palo Alto, Calif, Universidade Federal de São Paulo in São Paulo, Brazil, and Thomas Jefferson University Hospital in Philadelphia, Pa. Volunteer neuroradiologists annotated the images.

This dataset was created for the most recent RSNA Artificial Intelligence (AI) Challenge: to create an AI algorithm to assist in detecting and characterizing intracranial hemorrhage on brain CT. During prior challenges, researchers used existing datasets, but organizers wanted to create a new dataset for this one. The dataset creation project was a collaboration between the RSNA and the American Society of Neuroradiology. The AI challenge attracted at least 22,200 submissions from 75 countries, including 1,787 individual competitors in 1,345 teams.

ON-DEMAND WEBINAR

Brain Pathology Assessment with Panoramic Perfusion

Neuroradiology and AI in the news

A novel fully automated MRI-based deep-learning method for classification of IDH mutation status in brain gliomas. This study in Neuro-Oncology showed that researchers could use 3D brain images and a deep learning network to identify a specific genetic mutation in a glioma tumor with 97% accuracy. Currently, patients with these mutations require a pretreatment surgery to obtain glioma tissue for pathology, to choose appropriate treatment. The study, from UT Southwestern in Texas, aimed to identify status of the isocitrate dehydrogenase (IDH) gene, that can trigger brain tumor growth if the enzyme mutates. According to authors, this technique was able to determine the mutation status with one series of MR images, rather than multiple image types.

The need for a system view to regulate artificial intelligence/machine learning-based software as a medical device. This Nature NPJ Digital Medicine perspective piece from researchers at Harvard University and Europe-based INSEAD business school, argues that the U.S. Food & Drug Administration’s (FDA) proposal to regulate AI and ML, should factor in issues like physician training, and possibly only authorize that specific hospitals can use AI and ML products. This is counter to the FDA proposal to evaluate individual AI/ML products. The authors argue for using a system evaluation approach, which takes human factors into account.

Automated segmentation of acute stroke lesions using a data-driven anomaly detection on diffusion-weighted MRI. This study in the Journal of Neuroscience Methods discussed a novel and fully automated technique for detecting and segmenting acute ischemic stroke lesions on MRI scans, and classifying these lesions for stroke or non-stroke categories. This technique compared diffusion-weighted images (DWIs) and apparent diffusion coefficients (ADC) images, to a group of healthy images in voxel-level. The automated approach yielded good agreement with the expert manually-drawn lesions.

Promises of artificial intelligence in neuroradiology: a systematic technographic review. This study published in Neuroradiology was a systematic review of AI possibilities in neuroradiology, using an assessment of available AI applications during 2017-2019. It analyzed AI’s potential impact to neuroradiologists’ work. Authors concluded that AI in neuroradiology is not only available for clinical practice, but the functionalities mostly support radiologists. None of the studied applications completely replace radiologists but a few applications can take over the radiologist role for a limited set of tasks.

NICE publishes a Medtech Innovation Briefing on artificial intelligence for analyzing CT brain scans. The National Institute for Health and Care Excellence (NICE) reviewed software from five companies, using evidence from 11 studies, which automatically detected and notified healthcare professionals of abnormalities after CT brain scan analysis. It noted that the technologies might be of the most benefit when images can’t initially be reviewed by neuroradiologists.

As this newsletter shows, AI algorithms are increasingly gaining scientific validation in research and in the clinic. AI can supplement a radiologist’s work, providing high-quality data and sometimes alerting physicians to prioritize certain patients in the workflow. Increasingly, it’s important to incorporate these helpful tools in the healthcare environment.

Recent events have shown how important it is to have tools with high capacity and reliability. Artificial intelligence is one of the most promising technologies to meet these requirements and make any radiology department ready to meet the surge whether it was foreseen or not and becomes a go-to when it comes to finding new ways to improve current imaging and workflow practices in radiology.

Would you like to learn more about how AI works and how it can be implemented in your department, feel free to reach out to our experts!