“The brute-force methods that have worked so well in other domains . . . have not translated effectively to many parts of medicine . . .”
DeepHealth CEO Gregory Sorensen, M.D. on radiology AI’s annotated image challenges.
Imaging Wire Sponsors
- Arterys – Reinventing imaging so you can practice better and faster.
- Bayer Radiology – Providing a portfolio of radiology products, solutions, and services that enable radiologists to get the clear answers they need.
- GE Healthcare – Enabling clinicians to make faster, more informed decisions through intelligent devices, data analytics, applications and services.
- Healthcare Administrative Partners – Empowering radiology groups through expert revenue cycle management, clinical analytics, practice support, and specialized coding.
- Hitachi Healthcare Americas – Delivering best in class medical imaging technologies and value-based reporting.
- Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter.
- Riverain Technologies – Offering artificial intelligence tools dedicated to the early, efficient detection of lung disease.
- Siemens Healthineers – Shaping the digital transformation of imaging to improve patient care.
- Zebra Medical Vision – Transforming patient care with the power of AI.
The Imaging Wire
DeepHealth’s Progressive AI
A new study from RadNet’s DeepHealth subsidiary details how its forthcoming mammography AI algorithm could overcome many historical mammo AI challenges (limited annotated data, DBT support, generalizability), while detecting breast cancer 1 to 2 years before current practices.
- The Algorithm – The study focuses on an algorithm based on DeepHealth’s FDA-pending 2D/3D mammogram prioritization software package. The algorithm is reportedly able to achieve high accuracy with a limited volume of annotated images because of its unique training process that “mimics how humans often learn” by progressively training on more difficult tasks at each stage. They attributed the model’s generalizability to its variety of training sources (5 datasets, U.S., U.K., China).
- The Study – The researchers compared the model’s performance detecting breast cancer against five experienced breast radiologists. They used one set of 285 “index exams” performed 1-3 months before a biopsy-confirmed diagnosis (131 cancerous, 154 negative) and an earlier set of 275 initially-negative “pre-index” studies acquired 12-24 months before the index exams (120 cancerous, 154 negative).
- The Results – The algorithm outperformed all five radiologists with both the index set (+14.2% sensitivity, +24% specificity) and the pre-index exams (+17.5% sensitivity, 16.2% specificity), and would have identified almost 46% of the missed cancers at 90% specificity. The study also cited strong generalization with images from different sites, scanners, and populations that were not involved in the training set.
- The Takeaways – This study represents a solid start for DeepHealth’s forthcoming algorithm, which definitely would be well positioned if it can achieve these early detection results in clinical settings. It also introduces an interesting approach to AI training.
- FDA’s AI Action Plan: After several years of proposals and feedback, the FDA rolled out its healthcare AI regulation action plan. Here are its key actions and goals: 1) Create a new proposed regulatory framework; 2) Encourage best practices; 3) Support a patient-centered approach to AI development and transparency; 4) Support ways to evaluate and improve ML algorithms with a focus on bias and robustness; 5) Support pilots to understand how real-world evidence generation programs would work. These are some good goals, but it also shows how much further the FDA has to go before it is regulating AI the way the agency believes it should.
- Overnight Attendings and Recall Rates: When attending radiologists finalize ED reports during overnight shifts, recall rates can improve by at least 90%. That’s from a new University of Toronto study that found imaging-related ED recalls declined drastically in 2018 after they added attending radiologists to overnight shifts (7 recalls out of 15.1k overnight studies) compared to the previous two years when overnights were exclusively staffed by residents (2017 = 61 / 14.4k; 2016 = 54 / 13.8k).
- AI Metrics’ FDA: AI Metrics announced the FDA approval of its image analysis platform, which utilizes AI-assisted workflows to improve radiologists/oncologists’ accuracy interpreting CT and MRI scans to guide ongoing cancer treatments. The platform’s first application, AI Mass, supports the analysis and reporting of advanced cancer treatment.
- A PoC MRI Merger: Australia’s Magnetica (MRI engineering company) is set to acquire UK-based Scientific Magnetics (superconducting magnets company) and its U.S. subsidiary Tecmag (NMR, NQR, and MRI instrumentation). The combined company will develop lighter-weight superconducting MRI systems for point-of-care and extremity imaging.
- DL-US + BI-RADS Benefits: A South Korean research team developed a deep learning and BI-RADS-based breast ultrasound nomogram that can effectively differentiate masses and could help improve breast ultrasound’s false positive challenges. The nomogram uses ultrasound morphologic scores extracted from Samsung’s S-Detect deep learning software, combined with radiologists’ BI-RADS assessments and patients’ ages. Compared to radiologists’ BI-RADS assessments on their own, the nomogram achieved far superior tumor differentiation accuracy (0.87 vs. 0.51 AUC), false positive rates (45% vs. 97%), and false biopsy rates (48% vs. 98%), without affecting sensitivity (both 100%).
- Scatter Suppressed X-Ray: Polish scientists developed a ‘scattering suppression’ image processing algorithm that could lead to lower-dose and high-contrast X-rays, while eliminating the need for anti-scatter grids. By reducing X-ray scatter and the image-blurring noise that it creates, the technology could produce higher-quality images without higher-dosage X-ray.
- Redundant Research: A new JAMA paper revealed that many systematic review studies on COVID-19 are based on the same primary studies, suggesting that this level of duplication is “unjustified and may be unethical.” This comes after the team examined 25 systematic reviews on “imaging findings in children with COVID-19,” discovering that the reviews referenced just 17 primary studies with significant overlap (some used the exact same primary studies; a pair of primary studies were used in 19 and 20 of the 25 systematic reviews).
- TBI Blood Test: The FDA cleared an Abbott Laboratories blood test that evaluates whether patients with concussions require CT scans to check for tissue damage. The handheld device analyzes blood samples for TBI-related proteins, delivering results in 15 minutes. Abbott expects this to be the first test of its kind to gain widespread clinical adoption, noting that a previous FDA-approved test from Banyan Biomarkers never became widely available.
- Prostate MRI Underestimates: A UCLA study found that MRI frequently underestimates the size of prostate tumors, potentially leading to undertreatment. The researchers compared MRI-measured tumor size with actual tumor size after prostate removal (441 men, 461 lesions), finding that mean radiologic tumor sizes were much smaller than mean pathological tumor sizes (1.57 vs. 2.37). Underestimates were most significant with smaller radiologic tumor sizes and lower PI-RADSv2 scores, suggesting larger ablation margins may be necessary for these cases.
- ABR MOCs, Still Legal: Tennessee radiology Dr. Sadhish K. Siva’s second antitrust lawsuit against the American Board of Radiology (ABR) just reached a similarly unsuccessful result as his first. A U.S. district judge dismissed Dr. Siva’s latest argument that the ABR is creating a monopoly by tying its initial certifications to its ongoing maintenance of certification process (MOC). The ABR still faces plenty of challenges in the court of public opinion, but these last two judgements suggest that it is well positioned in the legal courts.
- Radiology’s Communication Gap: An ESR survey of over 400 patients across 22 EMEA countries found that although most patients are satisfied with the value of their radiological services (avg. rating = 4.22 out of 5), there’s room to improve patient communication and education. The survey revealed that 36% of patients were dissatisfied with the information they received about the risks and benefits of their scans and 33% weren’t satisfied with their radiologists’ availability to provide consultations.
- Big Imaging Acquisition Energy: It looks like 2021 could bring another wave of imaging center/practice acquisitions, as a PE Hub report suggests that Ohio-based radiology practice LucidHealth (PE-owned, 200 radiologists, 5 states) and large outpatient imaging center company SimonMed (physician-owned, 200 rads, 150 facilities, 9 states) are both on the market. The article listed a number of potential buyers (RP, CDI, US Radiology), while noting that LucidHealth and SimonMed are looking for substantial multiples over their respective 2020 EBITA levels ($60m & $75m).
- GI Ultrasound’s UC Advantage: A new study out of Australia found that gastrointestinal ultrasound could help identify which patients hospitalized with severe ulcerative colitis will not respond to corticosteroids, speeding up decisions to introduce salvage therapy. The researchers performed GI ultrasound on 10 consecutive severe UC patients within 24 hours of admission. The six patients who required salvage therapy had a 6.2mm median colonic bowel wall thickness (vs. 4.6mm), suggesting that patients with >6mm thickness in any colonic wall segment are likely to need salvage therapy. Meanwhile, endoscopy wasn’t able to identify which patients would require salvage therapy.
- Siemens Cios Flow’s FDA: Siemens Healthineers announced the FDA approval of its Cios Flow mobile C-arm, designed for a range of surgical disciplines (orthopedics, trauma surgery, spinal surgery, vascular surgery, and pain therapy). The Cios Flow is highlighted by its maneuverability, intuitive touch interface, and cost effectiveness, while launching with Siemens’ new SpotAdapt solution (helps visualize difficult anatomical areas) and new security functions.
The Resource Wire
– This is sponsored content.
- Learn how Windsong Radiology Group used Bayer’s MEDRAD Stellant FLEX CT injection system to drive cost efficiencies and standardization across its imaging centers.
- Learn how Hitachi Healthcare’s Sonticus system alerts physicians when critical findings thresholds are exceeded, helping them quickly address urgent issues and avoid diagnostic errors.
- The phrase “prevention is better than cure” was made famous 520 years ago, and it’s a core part of Zebra Med’s strategy today. Learn how and why Zebra Med is shifting its emphasis to population health as we head into 2021.
- Check out how Arterys’ Lung AI automates chest CT workflows, cutting reading times by 45% and reducing missed nodules by up to 70%.
- Siemens Healthineers detailed the six steps that radiology practices can follow to build resiliency, providing a roadmap detailing the workflow, leadership, and financial changes required to successfully respond to the pandemic and come out stronger because of it.
- Read about the 2021 CPT code changes impacting radiology practices in 2021 in this blog article from Healthcare Administrative Partners.
- Watch Jared Christensen, MD, MBA explain how Duke University Health uses Riverain Technology’s ClearRead CT Vessel Suppress and ClearRead CT Detect in its daily practice.
- This GE Healthcare Insight details why healthcare systems have to “work smarter, not harder” and how the company’s new “Intelligent Efficiency” approach can help make that happen.
- This case study details how University of Rochester Medical Center reduced its delayed diagnoses risk by 80% after adopting Nuance PowerScribe Follow-Up Manager.