- The State of AI: Nathan Benaich and Ian Hogarth just published their extremely comprehensive 2021 State of AI Report, detailing the latest in AI research, talent, commercialization, and politics, plus some future predictions. Imaging AI played a slightly smaller role in this year’s report than in 2020 (see slides: 37, 54, 55, 105), although the whole report is relevant if you’re in the AI space.
- POCUS Utility: Point-of-care ultrasound might allow clinicians to perform “visually satisfying medicine,” but there hasn’t been enough evidence-based studies to know how POCUS is affecting outcomes. That’s from a new JAMA paper calling for more research on POCUS’ role in patient treatment and clinical outcomes, and less studies focusing on POCUS’ usability, accuracy, efficiency, and cost.
- Philips Pediatric MRI Coaching: Philips introduced its new Pediatric Coaching solution, which combines augmented reality, gamification, and ‘buddy system’ techniques to reduce pediatric patients’ MRI-related anxiety. Patients begin interacting with Philips’ Pediatric Coaching characters from home using a mobile app, then interact with the same characters at the hospital while operating a child-size scanner, and finally they’re guided through the MRI exam by the characters’ images and audio projected within the MRI bore.
- Osteoporosis AI: South Korean researchers developed a hip X-ray DL system for predicting osteoporosis, suggesting that it could be a useful screening method in regions without access to DXA exams or for screening hip X-rays that were performed for other purposes. The team developed and tested the DL model using hip X-rays from 1,001 female patients (aged ≥ 55yrs, 504 w/ OP), detecting osteoporosis with 81.2% accuracy using an internal dataset and 71.8% accuracy with an external set.
- Subspecialty Benefits: Transitioning radiology workflows from a decentralized / modality-based structure to a centralized / subspecialty-based structure can make both radiologists and referrers more satisfied. A survey performed after a Swiss radiology network adopted the new centralized structure revealed that radiologists perceived their new reports to be higher quality (98.1% vs. 91.3%) and 95.7% of them preferred the new system. Meanwhile, referring physicians reported significantly higher overall satisfaction after the transition (94.3% vs. 84.9%) and viewed the new radiology reports as higher quality (92.4% vs. 72.8%).
- The Radiologist Label Advantage: Using radiologist-made image labels for AI training might not be scalable, but they’re more effective than using NLP-generated labels. A Singapore-based research team used 112k CXRs with radiologist and NLP-made labels to train three different pneumothorax detection models. Every radiologist-labeled model achieved significantly higher AUCs than the NLP-labeled models when tested against the official NIH dataset (0.881 vs. 0.838, 0.880 vs. 0.839, 0.943 vs. 0.869) and against CXRs from an external dataset (0.806 vs. 0.686, 0.871 vs. 0.736, 0.915 vs. 0.822).
- CMRI for Cardiac Tumors: A new European Heart Journal study provided the first large-scale validation of cardiac MRI’s accuracy diagnosing cardiac tumors, suggesting that CMRI could serve as a “one-stop-shop” for cardiac tumors. The multi-center study (n = 903 w/ suspected cardiac tumors) found that CMRI diagnosis matched clinicians’ final diagnosis with 98.4% of patients, while predicting mortality better than key clinical factors (LV ejection fraction, coronary artery disease, history of extracardiac malignancy).
- DeepSight’s $25M: DeepSight Technology announced a $25m Series A round that it will use to develop its novel ultrasound platform that it claims could achieve 100x higher sensitivity than current ultrasounds. Although details are limited, DeepSight’s ultrasound platform appears to be based around its sensor technology, which the company is combining with software and AI to achieve its goal of improving and expanding ultrasound imaging.
- Fast MRI’s Operational Impact: A new MGH study found that “fast MRI” significantly reduces acquisition times for the most common outpatient brain MRI protocols, providing both patient and operational advantages. A comparison of 1,000 exams before and after adopting fast brain MRI revealed that the accurately-named technique reduced MGH’s overall brain MRI acquisition times by 42% (18 vs. 31 min median), potentially allowing 10 more exams per day and $1.8m more reimbursements per scanner annually.
- Ultrasound’s Tape Alternative: A new study in the Surfaces and Interfaces journal found that using adhesive tape as an ultrasound probe cover could be as effective and more efficient (applied / removed faster) than current cleaning methods. The researchers used customized acrylate pressure-sensitive adhesive tape on a skin-like phantom, finding that the tape effectively prevents cross-contamination while maintaining “excellent” image quality.
- Delphinus’ SoftVue FDA: Delphinus Medical Technologies announced the FDA approval of its SoftVue 3D whole breast ultrasound tomography system as an adjunct to mammography for screening women with dense breasts (BI-RADS C or D). SoftVue can reportedly identify up to 20% more cancers in dense breasts than digital mammography, doesn’t require compression or radiation, and its exams’ unique prone positioning allows “an almost spa-like experience.”
