Are Automation and AI the Future of Brain Scan Analysis?
Complete the form below to unlock access to ALL audio articles.
After decades of development work, and trial and error and money spent, you might think that by the time your head is put inside a magnetic resonance imaging (MRI) scanner at a hospital, the difficult part of looking inside the brain is finished. But often it’s the analysis that comes after you leave the scanner that proves most challenging for physicians. The error rate for image analysis remains alarmingly high, and radiologists are being asked to handle and process larger numbers of scans every year. Can automation lend a hand? We talked to Dr. Chris Airriess, CEO at CorTechs Labs Inc., which has developed a post-processing scan software called NeuroQuant designed to streamline the analysis pipeline. We asked Chris about the current challenges in analysis, what datasets NeuroQuant would be trained on, and whether the general public will trust their medical scans to an AI.
Ruairi Mackenzie (RM): What is the current workflow for analyzing a neuroimaging scan? What is difficult now? What are the primary concerns of docs?
Chris Airriess (CA): Clinicians are more pressed for time than ever with mountains of data to analyze and a workflow that never stops. When it comes to analyzing radiological images, radiologists must form their assessment as quickly as possible – a major challenge in subtle cases with noisy or low-resolution scans and a lack of objective empirical evidence to support a diagnosis. Though such decisions are based on their best judgment, another physician may have a different assessment, or the same physician may assess the same scan differently if reviewed on another day. By post-processing the scan with software such as NeuroQuant, radiologists can use objective numeric data to identify and track many clinically relevant biomarkers.
Magnetic resonance technology has improved greatly since its adoption for clinical use in the 1980s and often plays a crucial role in aiding physicians to diagnose and monitor many diseases and conditions. As with all technologies, the ongoing evolution of MRI techniques is continuous, with the aims of improving patient comfort and optimizing the pathway to diagnosis. With today’s artificial intelligence and computer-based algorithms, image post-processing is playing an increasingly important role in providing objective measurements and biomarkers so physicians need to order fewer tests before making a definitive diagnosis.
RM: How can automation improve that workflow?
CA: Automation solves workflow challenges by increasing efficiency and reducing human error and may also increase uniformity across readers in a practice. Specifically, automation may help draw attention to regions of pathology that might not be caught with a naked eye read, provide objective empirical evidence to support a diagnosis, reduce vague impressions and may lower the number of tests required to make a definitive diagnosis. Automating workflows also improves the speed of reading longitudinal studies with automatic comparisons providing precise quantification of change as well as visual highlights of changing regions. It also reduces dictation errors, by auto-populating reports with data when software are integrated with dictation systems. Automation also provides objective data that may decrease the need for over-reads or second interpretations.
RM: How will data analytics-powered neuroimaging improve the diagnosis of neurological disease?
CA: Data analytics-powered neuroimaging is a logical next step, where stakeholders have great hopes of streamlining the diagnosis of neurological conditions. Patients will benefit by receiving an earlier, more accurate diagnosis, and radiologists will benefit from objective data for interpretation, higher throughput and the ability to provide more meaningful reporting to their referring physicians.
A key strength of neuroimaging powered by data analytics is that it can provide empirical evidence in subtle cases, eliminating guesswork and vague impressions. It also improves patient care management (such as evaluating response to drugs, therapies and treatments) and improves patient outcomes by getting patients the correct treatments and tracking treatment responses over time.
RM: AI is only as good as the data that it is trained on. What datasets will CorTechs Labs’ software be trained with?
CA: NeuroQuant has been developed and trained on thousands of curated cases from public and private sources. The normative database was established from several thousand curated scans, including publicly available studies, collaboration partners and other proprietary data sources covering the age range from three to 100.
RM: Would medical professionals still have a role to play during image analysis in an automated workflow?
CA: Yes. Automated workflow is designed to assist medical professionals in their overall impression and decision making, so they can focus their time on higher-level analyses rather than mainly manual and subjective processes. Additionally, in order to be reimbursed by insurance or Medicare, a radiologist must review all data. Physicians must understand the relationship of the data to their patients and perform a visual quality control check of all images. Automation software is designed to assist - but not replace - medical professionals.
RM: Will patients trust AI to assess their medical images instead of their doctor?
CA: An increasing number will in time, as AI improves and becomes more ubiquitous throughout their everyday lives. Soon, I expect that patients will request that their physicians employ AI tools, because the benefits will become common knowledge, and everyone wants the best patient care. To reinforce my earlier statement, this technology is here to aid physicians in their analysis of data and to reduce human error where possible, doctors work hand-in-hand with the technology to provide the best possible diagnosis, treatment and follow-up monitoring.
RM: Is analyzing the connectome the future of neuroimaging?
CA: Bringing MR diffusion tractography into the clinical workflow will, along with several other emerging technologies, greatly enhance the abilities of neuroradiologists to provide meaningful insights to referring physicians and ultimately to their patients. The connectome is getting a huge amount of attention at most neurologically-focused conferences these days, and a proliferation of large, well-funded research studies are exploring how neural circuitry develops normally as well as how things break down. New technical capabilities in this area will continue to advance and accelerate our knowledge of how the brain works on both structural and functional levels.
Dr. Chris Airriess was speaking to Ruairi J Mackenzie, Science Writer for Technology Networks