#404 – The Wire

  • Amateur Ultrasound AI: Northwestern Medicine and Google are teaming up to develop a handheld ultrasound AI solution intended to help expand access to fetal ultrasound interpretations in low and middle-income countries. Since these exams would be performed by lightly trained health workers or even pregnant women, Northwestern and Google will compile a training database of “amateur” handheld ultrasound images (performed by Northwestern patients, family members, and novice clinicians), matched with professional images and diagnoses. Google Health will then use this database to develop an AI solution to accurately interpret amateur fetal ultrasound images.
  • CDIs PCa Advantage: A University of Waterloo-led research team showed that Synthetic Correlated Diffusion Imaging (CDIs) could improve MRI prostate cancer screening. The researchers performed CDIs and standard prostate MRI techniques (T2w, DWI, DCE) on 200 patients, finding that CDIs differentiated cancerous and healthy tissue (0.916 AUC) and clinically significant and insignificant prostate cancer (0.755 AUC) far more accurately than the standard MRI techniques.
  • Is Pricier Healthcare Worth It? A recent study from the National Bureau of Economic Research found that US hospitals with higher costs of care have lower overall mortality rates, with a 35% mortality difference between hospitals in the 20th and 80th cost percentiles. However, the reduction was driven by hospitals in competitive markets where patients have alternatives, while receiving care from pricier hospitals in less competitive markets where competition isn’t geographically feasible (reportedly 69% of all hospitals) has no detectable effect on mortality.
  • Arterys & RMS’ BENELUX Alliance: Arterys and Belgium-based medtech developer and distributor, RMS Medical Devices, launched a partnership that will make Arterys’ AI platform available through RMS in Belgium, Netherlands, and Luxembourg. The alliance continues Arterys’ notable European channel expansion since the start of 2021, following partnerships with AMG Medtech (UK & Ireland), Wellbeing (UK), and Human Bytes (Nordic countries). 
  • Contrast Incident Reporting: A new JACR study detailed MGH’s successful rollout of a tool that supports radiologists’ imaging contrast reaction documentation process (CISaR; Contrast Incident Support and Reporting). Analysis of 431 CT and MRI contrast reactions from before and after MGH adopted CISaR showed that it significantly increased the share of reactions with radiologist documentation (from 50% to 89%) and documentation completeness (>95% reports complete), while achieving full adoption within nine months of its introduction.
  • Off Label AI: A team of researchers from UC Berkeley and University of Texas revealed that images in many open-source databases were preprocessed to support specific AI use cases, and therefore are likely to produce unexpected results when used with other AI models. The authors realized this after they were unable to replicate an image reconstruction study that used a preprocessed MRI dataset. They then performed a number of tests that revealed MRI image reconstruction algorithms can produce 25% to 48% higher quality images when they use preprocessed images (vs. raw).
  • SPADE Ultrasound: Researchers from The Technion – Israel Institute of Technology unveiled their new ‘SPADE’ ultrasound detector technology (Silicon-Photonics Acoustic Detector), which leverages optical components to support smaller ultrasound systems with better image quality (can be built into 1 mm devices, can image 10-20 micrometer objects). The new SPADE detectors could support new endoscopic and vascular ultrasound applications, noting that the extremely small transducers required for these use cases traditionally have low image quality.
  • PE AI with Low Quality CTPAs: A new MGH study showed that Aidoc’s pulmonary embolism AI solution maintains its accuracy with both suboptimal and optimal quality CTPA images (n = 104 & 226), detecting PE with similar sensitivity (100% vs. 96%), specificity (89% vs. 92%), and AUCs (0.89 vs. 0.87). Although it’s rare to see AI studies focused on performance with low-quality exams, this could be an important capability considering that another recent study showed that the same Aidoc solution was particularly helpful catching missed PE cases in CTPAs with lower contrast injection quality.
  • 177Lu-PSMA-617 Cleared: The FDA cleared Novartis’ 177Lu-PSMA-617 (Pluctivo) prostate cancer treatment for patients with PSMA–positive metastatic castration-resistant prostate cancer (mCRPC). The approval should be welcomed by these patients and their care teams (reduces risk of death by 38% and progression by 60%) and should lead to more PET scans given that it relies on PET to identify and treat patients.
  • Prioritizing Abnormal Brain MRIs: A King’s College-led team developed a deep learning-based framework for triaging and prioritizing brain MRIs, finding that it could significantly reduce abnormal case reporting times. The team trained the model with 70k labeled T2 and axial diffusion-weighted MRIs and tested it against 800 MRIs, finding that it was able to quickly (<5 seconds) and accurately (>0.9 AUCs) identify abnormal brain MRIs, with good generalization across the two hospitals networks (ΔAUC ≤ 0.02). A simulation study showed that the model would have significantly reduced the two networks’ average abnormal exam reporting times (28 days to 14 days & 9 days to 5 days).

You might also like

You might also like..

Select All

You're signed up!

It's great to have you as a reader. Check your inbox for a welcome email.

-- The Imaging Wire team

You're all set!