|
Detecting the Occult | Cancer Blood Test
December 20, 2021
|
|
|
|
|
Together with
|
|
|
|
“Training, testing and validating on the same dataset is so 2020.”
|
|
A Twitter comment from CARING Research’s Vidur Mahajan regarding the aortic stenosis AI study detailed below.
|
|
|
|
A new study published in European Heart Journal – Digital Health suggests that AI can detect aortic stenosis (AS) in chest X-rays, which would be a major breakthrough if confirmed, but will be met with plenty of skepticism until then.
The Models – The Japan-based research team trained/validated/tested three DL models using 10,433 CXRs from 5,638 patients (all from the same institution), using echocardiography assessments to label each image as AS-positive or AS-negative.
The Results – The best performing model detected AS-positive patients with an 0.83 AUC, while achieving 83% sensitivity, 69% specificity, 71% accuracy, and a 97% negative predictive value (but… a 23% PPV). Given the widespread use and availability of CXRs, these results were good enough for the authors to suggest that their DL model could be a valuable way to detect aortic stenosis.
The Response – The folks on radiology/AI Twitter found these results “hard to believe,” given that human rads can’t detect aortic stenosis in CXRs with much better accuracy than a coin flip, and considering that these models were only trained/validated/tested with internal data. The conversation also revealed a growing level of AI study fatigue that will likely become worse if journals don’t start enforcing higher research standards (e.g. external validation, mentioning confounding factors, addressing the 23% PPV, maybe adding an editorial).
The Takeaway – Twitter’s MDs and PhDs love to critique study methodology, but this thread was a particularly helpful reminder of what potential AI users are looking for in AI studies — especially studies that claim AI can detect a condition that’s barely detectable by human experts.
|
|
|
Nanox.AI Thinks Big
See how and why Nanox.AI sees a much bigger future for public health AI than many of us imagine in this Imaging Wire Q&A with Zohar Elhanani.
|
|
GE’s Three Steps to Thrive Tomorrow
Ready to act today in order to thrive tomorrow? This GE Healthcare report details the three steps that healthcare institutions can take to improve productivity, effectiveness, efficiency, and patient outcomes.
|
|
- Lung Cancer Blood Test: MGH researchers developed a blood test that can identify early-stage lung cancer in asymptomatic patients, potentially serving as an initial screening step before CT imaging. The blood metabolites-based predictive model was able to identify 25 patients with non-small cell lung cancer from thousands of specimens. In a later test, it identified 54 patients with NSCLC using blood samples obtained before their diagnosis, suggesting that its early predictions are accurate.
- A Developing World MRI: A team of Hong Kong-based researchers produced a low-field MRI (0.055T) that’s easy to install (compact, low-maintenance, uses standard AC power outlet, doesn’t require RF shield or cage) and could be manufactured for under $20k, making it ideal for developing countries. The scientists used the MRI to perform four common brain imaging protocols on 25 patients (T1 & T2, FLAIR, DWI), showing that it could be used to diagnose brain tumors or stroke.
- Motilent’s Gut Imaging 510k: Gut-imaging AI startup Motilent announced its entry into the US market, backed by its GIQuant software’s new FDA 510k and an AI marketplace alliance with Nuance. GIQuant is an MRI post-processing solution that measures/tracks intestinal movement to provide insights into gastrointestinal disease progression and treatment effectiveness (initially targeting Crohn’s Disease).
- Stand-Alone AI: A new study out of Spain detailed a stand-alone AI system that can interpret digital mammography and DBT screening exams as well as radiologists, potentially eliminating 90% of rads’ breast cancer screening workload. The researchers used ScreenPoint Transpara to read 16k DM and DBT exams (w/ 98 screen-detected cancers, 15 interval), achieving 0.93 and 0.94 respective AUCs, and noninferior sensitivity to radiologists when used as a single or double reader. The AI tool reduced recall rates by 2% with DM exams, but had a 12% higher recall rate with DBT exams.
- Signify’s RSNA Impressions: Signify Research published its annual RSNA overview, highlighting the show’s biggest trends (focus on efficiency, high funding & competition, looming consolidation, digitalization’s maturation) and detailing the major takeaways within imaging’s key segments. Here’s some of those RSNA takeaways: AI (larger players becoming more dominant, focus on care coordination, more AI partnerships, minimal consolidation… so far), imaging IT (more focus on productivity and operational outcomes), modalities (big CT and MRI innovations, focus on efficiency/usability, growing role of AI image enhancement, limited X-ray and ultrasound activity).
- Believing in Black Boxes: A recent University of Toronto editorial argued that healthcare machine learning tools don’t need to be explainable to be evidence-based. The authors believe explainability isn’t essential, noting medicine’s long list of treatments / procedures that are not fully understood (e.g. acetaminophen or gastric bypass surgery). Instead of holding ML to unnecessarily high standards, the authors suggest that we should focus on improving how we evaluate ML, an approach that aligns with another recent paper from some of our favorite AI thinkers.
- Walgreens Primary Care: Walgreens revealed plans to operate primary care clinics at nearly half of its 9k stores, staffing two physicians at each clinic (that would be 5% of the US PCP workforce). With CVS also planning to staff physicians in its upcoming primary care locations, it appears we might be heading toward a future where retail clinics account for a significant portion of the primary care landscape.
- Brainomix Branches Out: UK-based stroke AI startup Brainomix closed a $21.2M Series B round (total now $34.9M) that it will use to fund its expansion into new clinical indications (lung fibrosis and cancer) and new markets (drug discovery).
- First Photon Counting Evaluation: Researchers from Mayo Clinic and Siemens Healthineers released the first technical performance evaluation of Siemens’ NAEOTOM Alpha photon-counting CT, helping to confirm the industry’s first PC-CT’s imaging advantages. Test exams revealed that the NAEOTOM Alpha achieved 125-micron in-plane spatial resolution and 0.3 mm longitudinal resolution in high-resolution mode (the smallest reported from a clinical CT), and 66-msec temporal resolution multi-energy imaging in dual-source mode, while significantly reducing noise (up to 47%) or dosage (up to 30%) compared to conventional CTs.
- ADMdx CorInsights MRI: ADM Diagnostics (ADMdx) announced the launch of its CorInsights MRI software (FDA cleared), which measures regional brain tissue volumes using standard MRI scans, compares volume data against a reference database, and uses the comparisons to produce patient-specific reports.
|
|
Magnolia Regional and Nuance Catch More Cancers Sooner
Discover how Magnolia Regional Health Center started catching more cancers sooner when it adopted Nuance’s PowerScribe Lung Cancer Screening Program and PowerScribe Follow-up manager.
|
|
- The USPSTF guidelines for lung cancer screening were updated in May 2021, and driving compliance to such guidelines is a long, slow, repetitive process. Because of that, the Riverain team put together a kit to help hospitals and imaging centers educate either referring physicians or patients on the new guidelines either via branded tools or through the media.
- Check out this Imaging Wire Q&A, where Arterys CEO John Axerio-Cilies, PhD discusses medical imaging’s AI and cloud evolution and how Arterys works with its Center of Excellence partners to make AI real.
- Trying to grow your imaging practice? See the strategies Desert Radiology used to manage its growth in this on-demand webinar from Aunt Minnie and United Imaging.
- With radiation dose management now largely considered best practice, this Bayer white paper details the top five benefits of adopting contrast dose management.
- Despite significant interest, there’s still confusion about the value of imaging AI. This Blackford Analysis white paper explores the key cost considerations and ROI factors that radiology groups can use to figure out how to make AI valuable for them.
|
|
|
|
|
