- Smartphone Stroke Screening: New research shows that smartphone videos paired with video-based motion analysis (VMA) could be “excellent” for carotid artery stenosis screening (CAS; usually done w/ ultrasound). The researchers captured 30‐second smartphone videos of 202 patients’ necks (54% w/ CAS) and then used VMA to quantify skin motion changes, identifying CAS with a 0.914 AUC. This technique joins an emerging list of imaging-adjacent smartphone diagnostics, which also includes concussions/TBI (patient eye videos) and COVID (patient cough audio).
- United Imaging’s Big IPO: United Imaging Healthcare made its debut today on Shanghai’s STAR Market exchange as part of a massive $1.62B IPO, which equaled a 78x-multiple of UIH’s earnings (STAR’s average is 35x) and became STAR’s biggest IPO of 2022. United Imaging’s influx of public capital will be directed towards its R&D, manufacturing, and marketing operations, suggesting that the already-disruptive OEM is about to become even more aggressive.
- Evaluating AI Value Propositions: A new European Radiology paper provided a detailed and refreshingly visual breakdown of how imaging AI vendors position and legitimize their solutions. Analysis of 393 AI apps from 133 vendors found that most AI value propositions focus on quality-of-care (31%), efficiency (18%), or both quality and efficiency (28%). AI vendors’ efforts to legitimize these messages focused on their clinical / research partnerships (75%), regulatory approvals (72%), team expertise (56%), and clinical implementations (53%). However, very few vendors revealed the sources or size of their training datasets (8% & 10%).
- Pro Medicus’ Plans: On the heels of Pro Medicus/Visage Imaging’s “most successful” fiscal year ever (revenue +38%, net profit +44%), CEO Sam Hupert shared new insights into the company’s future strategy. Hupert highlighted Pro Medicus’ ample remaining growth in the US (still has ~5% share, can expand beyond academic centers), while suggesting that an economic downturn could make acquisitions a larger part of its strategy (target adjacent imaging startups… like AI, not PACS competitors). Meanwhile, Pro Medicus’ R&D efforts will continue to prioritize expanding to “other ologies,” including adding cardiology to its existing imaging platform.
- Demographics and Mammography Perceptions: A Duke study highlighted the major influence demographics have on how patients perceive their mammography screening reports. Out of 178 women (71% White), Black patients were less likely to be satisfied with report quality (p=0.043), but more likely to trust their report’s findings if their radiologist was also Black (p=0.037). Meanwhile, participants without any college education were less likely to be satisfied with their report quality (p=0.020) or feel that their radiologist cares about his/her patients (p=0.037).
- MediMatrix Acquired: Mobile imaging software company MediMatrix (formerly WebInterstate) was acquired by ASG, a PE-backed company that’s assembling a portfolio of SaaS companies in targeted verticals (including healthcare). MediMatrix develops software used in mobile imaging operations (technologist dispatch, order management, image sharing, billing), which seems to have gained new momentum with the current home care trend. ASG plans to expand MediMatrix’s business organically and through future healthcare acquisitions.
- Brain Tumor MRI AI: A new JAMA study detailed an MRI-based AI model that improved classification and diagnosis of 18 types of brain tumors, both when used independently or by neuroradiologists. The researchers trained and tested the AI model with MRIs from 37.8k and 1.3k patients, finding that it outperformed nine neurorads (accuracy: 73.3% vs. 60.9%; sensitivity 88.9% vs. 53.4%; specificity 96.3% vs. 97.9%). Perhaps more importantly, neurorads who used AI were more accurate than neurorads without AI (75.5% vs. 63.5%).
- Imaging Heart Disease in Women: A review in JACC highlighted the need to consider women’s unique cardiac imaging features and disease physiology. The authors emphasize that women have a unique ischemic heart disease phenotype (less calcified lesions, more nonobstructive plaques, higher prevalence of microvascular disease) and that imaging tests tend to be less accurate with women (summarized wonderfully in this image).
- COVID Time-to-Death AI: Japanese researchers developed a chest X-ray and clinical data-based AI model that accurately predicted COVID patients’ likelihood of dying within days of their admission. The researchers combined a DeepSurv model (based on the Cox hazards model) and a deep learning CNN, and trained and tested it using data from 1.35k COVID inpatients. The multi-modal AI system predicted time-to-death more accurately than versions of the same AI model that used only CXR images, only clinical data, or only the Cox proportional hazards model (c-index 0.82 vs. 0.77 & 0.70 & 0.71).
- Scientific Misconduct: Authors of an Academic Radiology study identified 192 retracted medical imaging papers that were published between 1984 and 2021, noting a steady rise in retractions since 2000. The retracted articles were most often published by teams based in China, the US, Japan and South Korea (31.3%, 12.5%, 7.3% and 6.3%), and were most commonly pulled due to duplication of other papers, plagiarism, and data concerns (7.1%, 6.8% and 5.4%) – with “scientific misconduct” identified in 55.7% of the retracted articles.
- Considering Radiation Risks: A new study in the European Journal of Radiology found that most referring physicians consider radiation risk when ordering CT scans, and they’re open to regulations that would control the number of CTs patients receive annually. In an email survey of 505 referring physicians from 24 countries, 58% understood that current regulations do not limit patients’ annual CT volumes, but 69% are open to regulations controlling CT volumes (36% “maybe,” 33% “yes”).
