Uncategorized

Multimodal CXR | RadNet’s AI Plan | CTC Opportunity

“this is perhaps the biggest opportunity that our industry and our company has seen in its history.”

RadNet CFO Mark Stolper on AI.



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.
  • Novarad – Transformational imaging technologies that empower hospitals and clinicians to deliver clinical, operational and fiscal excellence.
  • 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.
  • United Imaging – Our mission, Equal Healthcare for All, pushes us beyond conventional boundaries to help clinicians expand modern, digital, intelligent care to more people within their communities.
  • Zebra Medical Vision – Transforming patient care with the power of AI.




The Imaging Wire


Multimodal X-Ray

IBM Research just released a multimodal chest X-ray dataset that could represent a major advancement from the datasets we use today.

  • The Multimodal Dataset – The 1,083-image dataset combines CXR images, radiologist eye-tracking recordings, localized disease labels, report text, and radiologist dictation audio – and the eye-tracking and dictation recordings are synced.
  • AI Model Opportunities – Multimodal datasets like this could allow AI developers to create more sophisticated and explainable AI models, while potentially leading to models that are better aligned with radiologists’ actual diagnostic processes.
  • Labeling Opportunities – This approach could also help build single-modal imaging databases, as its combination of eye-tracking and disease localization could be used to create automated image labeling workflows (during interpretations).
  • Next Step: Grow the Dataset – The researchers made their dataset and processes open-source, encouraging other groups to create similar datasets and help drive multimodal AI forward.



RadNet’s AI Plan

While the debate over how AI will affect radiologists’ careers continues to play out in academic papers, RadNet just made it’s plan for how AI will affect the business of radiology far less debatable.

  • AI Timeline During a recent investor conference presentation, RadNet’s CEO forecast that all of its mammograms will be read by AI within “several years, if not sooner,” helping the major imaging network improve efficiency and accuracy (fast forward to minute 18).
  • About AI Efficiency – RadNet was sure to clarify that it’s not planning to replace breast radiologists with AI. However, RadNet is definitely looking to cut its mammography interpretation costs with AI, noting that 20% of its revenue goes to paying radiologists (~$200m) and mammography interpretations represent one-fifth of its radiologist costs (~$40-$50m).
  • Alarming, But Not Surprising – These statements may be alarming to radiologists who are concerned that AI will cut their earnings (that’s literally RadNet’s plan for mammography). However, they shouldn’t be surprised, given that RadNet has actively sought out AI acquisitions (DeepHealth, Nulogix) and partnerships (Hologic, WhiteRabbit.ai) over the last two years. It also wouldn’t be surprising if other PE-backed radiology groups have similar plans.

The Wire

  • The Recommendation Effect: When the U.S. Preventive Services Task Force (USPSTF) added CT colonography (CTC) to its colorectal cancer screening guidelines in 2016, CTC rates jumped by 50%. That’s from a new Emory-led study that reviewed private insurance claims from 2010–2018 and found that monthly CTC rates jumped from 0.4 per 100k people before 2016 to 0.6 per 100k people after the recommendation.
  • Thirona’s Cystic Fibrosis AI: Thirona just launched perhaps the first commercial AI tool used to detect and quantify lung abnormalities related to cystic fibrosis. Thirona’s new PRAGMA-AI algorithm analyzes CT scans to automate PRAGMA-CF cystic fibrosis quantification method (takes several hours per patient, requires trained data analysis). PRAGMA-AI will be included in Thirona’s LungQ software package that’s already available in Europe, the US, and Australia.
  • Explaining the UK’s Lung Cancer Delays: A Cancer Research UK survey (n = 1,000 primary care docs) attributed the UK’s massive drop in lung cancer referrals and treatment starts during the COVID pandemic (-34% & -9% declines) to patients’ reluctance to go into hospitals for tests or even to disclose their symptoms (91% & 78% of responses). The PCPs also blamed the declines on diagnostic test backlogs, telehealth consultation challenges, and long COVID-19 test turnaround times prior to diagnostic lung cancer exams (73%, 68%, and 54% of responses).
  • CT’s CTC Opportunity: A new MGH study detailed how healthcare providers could improve colorectal cancer (CRC) screening rates by performing CT colonography (CTC) scans during patients’ CT exams. A survey of 16.2k adults (45-75yrs, no history of CRC) revealed that 49.2% of the respondents have not received CRC screening, while 33.7% of those same respondents did undergo a CT scan. Based on that math, the researchers estimate that 15.2m people in the US could have been brought up-to-date with their CRC screening if they received a CTC scan during their CT exam.
  • Philips Focuses: Philips will sell its home appliance business to PE firm Hillhouse Capital for €4.3 billion (including ongoing brand licensing fees), allowing Philips to focus on its personal and professional health businesses. Philips first announced plans to sell its appliance unit over a year ago (and it wasn’t a major surprise back then), but this is still a notable milestone and it continues a trend of the major imaging OEMs becoming more healthcare focused.
  • AR-PAM/US for CRC: Researchers from Washington University in St. Louis developed a new multimodal imaging technique that could significantly improve outcomes for patients treated for colorectal cancer. Their new AR-PAM/US method combines photoacoustic microscopy with ultrasound to capture rectal tissue images, and then uses deep learning to identify residual cancers not eliminated by chemotherapy and radiation treatment (more accurately than MRI). The team suggests that AR-PAM/US could help identify patients who responded well to chemoradiation, avoiding unnecessary and sometimes harmful surgeries.
  • Risk-Based Mammography Prioritization: A new JAMA Open study found that using breast cancer risk levels to prioritize mammography screening appointments could be valuable during periods of decreased capacity (e.g. a post-COVID surge). The cohort study (898k women, 1.878m mammograms) found that women with high-to-very high breast cancer risks represented 12.1% of all mammograms but 55% of all breast cancers. Conversely, women with low breast cancer risks represented 44% of mammograms and just 13% of detected cancers.
  • Senate OKs Medicare Cut Delay: The U.S. Senate overwhelmingly passed (90-2) a bill to further postpone automatic 2% Medicare spending cuts until the end of 2021. The bill is now heading to the U.S. House for final passage. Congress originally delayed the “automatic” 2% cuts as part of the CARES Act and postponed the cuts several times since then.
  • NLP Incidental Pipeline: A Duke University team developed an NLP pipeline that reviews chest CT reports to identify incidental thyroid nodules (ITNs) that require follow-up ultrasounds. The team developed their fastText NLP tool using >11.5k non-contrast chest CT reports from 2018 and tested it against >10k reports from 2017, identifying the 81 ITNs that required follow-up with 96.5% accuracy. Given that only 38 (47%) of these reports actually recommended follow-up ultrasound exams, an NLP pipeline like this could help improve ITN follow-ups.
  • The COVID CT Curve: A new NIH study demonstrated how CT dynamic curves can help physicians monitor COVID progression and spot patients at risk of deteriorating. The team used an NVIDIA-based algorithm to analyze 121 patients’ CT scans and laboratory results to produce a CT dynamic curve showing that: 1) Lung opacities appeared an average of 3 days before symptoms started and peaked one day later; 2) Ground glass opacities generally appeared before consolidation, and persisted after consolidation resolved. Physicians could compare patient CTs against this progression curve to identify and treat more severe outliers.
  • Riverain ClearRead CT’s VA Expansion: Riverain Technologies’ ClearRead CT lung nodule detection software further expanded within the U.S. VA system after gaining an award for Veterans Integrated Service Networks 8 (VISN 8, Florida, S. Georgia, Puerto Rico). VISN 8 is the twelfth VISN to use ClearRead CT out of 18 total VISNs.
  • SubtleGAD Evidence: A new study out of China confirmed that Subtle Medical’s SubtleGAD solution effectively supports reduced-dose brain MRIs. The researchers used contrast-enhanced brain MRIs from 83 consecutive patients (30 for DL training, 53 for testing), comparing three 3D T1-weighted images from each patient (0% dose, 10% dose, and 100% dose). The SubtleGAD-enhanced images accurately matched the full-dose images in 48 of 53 cases (90.6%), catching 34 of 36 cases that had a single lesion (94.4%). However, the SubtleGAD-enhanced scans performed worse among patients with multiple lesions (identifying all lesions in just 3 of 6 cases), most commonly missing these patients’ smaller lesions.
  • The Virtual Nodule Clinic: Optellum announced the FDA clearance of its Virtual Nodule Clinic, an AI-powered clinical decision support tool that pulmonologists and radiologists can use to identify and track patients with suspicious lung nodules. The Virtual Nodule Clinic uses CT radiomics to produce Lung Cancer Prediction (LCP) scores used to support diagnosis and monitoring decisions.

The Resource Wire

– This is sponsored content.

  • It’s clear that structured reporting is a must for CVIS platforms, but they aren’t all created equal. This Hitachi article reveals what physicians and sonographers view as the “non-negotiable” CVIS structured reporting features.
  • On March 21st, United Imaging celebrated its tenth year in business. Since its founding in 2011 the company has over 13,000 global installations and continues to grow. A video on LinkedIn celebrates United Imaging’s employees and customers worldwide.
  • This Healthcare Administrative Partners post details how radiology practices will be affected by the upcoming national surprise billing rules and outlines how they should prepare for these changes.
  • Novarad’s COVID-19 AI Diagnostic Assistant analyzes chest CT scans in seconds, helping physicians quickly identify COVID-19 patients and get them into care. The best news – it’s available to clinicians worldwide free of charge.

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!