Uncategorized

Delivery Science | CAD Faceoff | Brain Maps

“We’re treating sets of modalities almost like a musical note,”

Georgia State University professor Vince Calhoun on his team’s efforts to identify which modalities and types of imaging data (in this case, the notes) could be combined (making them cords) in order to map brain disorder patterns.


Imaging Wire Sponsors

  • Bayer Radiology – Providing a portfolio of radiology products, solutions, and services that enable radiologists to get the clear answers they need.
  • GE Healthcare – Providing point of care ultrasound systems, from pocket-sized to portable consoles, designed to support your clinical needs and grow along with your practice.
  • 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.
  • 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.

The Imaging Wire


Delivery Science

There’s no doubt folks are excited about healthcare AI, and the ever-growing list of AI models and funding rounds are prime examples of that. However, a new paper out of Stanford suggests that healthcare AI will only be able to achieve scalable and sustained clinical value through major improvements in AI delivery.

  • The AI Impact Gap – To support their argument, the authors pointed out the gap between AI development efforts and real world adoption, suggesting that AI will continue to live on healthcare’s “innovation” margins until AI delivery science catches up.
  • AI Delivery Science Principles – The authors proposed that AI delivery should be based on the understanding that: 1) Much of healthcare is delivered via complex systems and AI must accommodate this complexity; 2) AI should be viewed as an enabling component of broader solutions, not as the end product; and 3) AI-enabled complex systems include a combination of people, processes, and technologies. Therefore, AI implementation shouldn’t just be about deploying a model, it should be about designing the best possible care delivery system for a given problem, with the AI playing a role in that system.
  • How to Catch up – The paper suggests that AI delivery science will require a broader set of tools (design thinking, process improvement, and implementation science) and a broader definition of how AI-integrated healthcare works (a combination of AI models and AI-connected care delivery systems). AI developers will also have to evolve, with a strict focus on aligning with clinical use cases (not just getting accurate predictions) and potentially requiring AI development to shift from the virtual world to actual healthcare environments.
  • The Takeaway – There’s no shortage of papers about healthcare AI’s challenges, but the majority of those critiques target the technical side of AI (brittle solutions, narrow AI, black box validation, annotation limits). Those are all still important problems to solve, but the AI industry’s fixation on these very software-focused issues might be the most compelling evidence that we need to focus more on AI delivery.



CAD Faceoff

Swedish researchers published the first study comparing multiple commercially available mammography CAD solutions, finding that one of the solutions outperformed radiologists. Perhaps more importantly, they found that by combining the top-performing CAD system with a first-reader radiologist, they identified more cancers than by combining first- and second-reader radiologists.

  • The Study – The researchers used three commercially available CAD systems to read screening mammograms from 8,805 women (40-74yrs, 739 diagnosed w/ breast cancer in 12mo), both independently and in combination with human radiologists.
  • The Results – The three CAD systems identified 81.9%, 67%, and 67.4% of the breast cancer cases, while first-reader radiologists identified 77.4% of cases and second-reader rads identified 80.1% of cases. The combination of the top CAD system with first-reader radiologists, identified even more breast cancer cases (88.6%) than a process using human radiologists as first and second readers.
  • Significance – The main takeaways from this study might depend on the observer’s industry role. To breast health advocates, this is more evidence that AI could improve breast cancer screening effectiveness and expand screening to underserved regions. To AI developers and buyers, its evidence of how different CAD performance can be. To researchers, it suggests that prospective clinical studies could use the top-performing CAD systems as independent readers.

The Wire

  • Structured & Automated TI-RADS: A Duke University study in JACR found that using a structured and automated thyroid ultrasound reporting template based on ACR TI-RADS significantly reduces errors and dictation times. In the study, four radiologists dictated thyroid ultrasound reports using four different TI-RADS templates. The structured and automated reports had zero errors (0/80), which was far less than the free text (27.5%, 22/80), minimally structured (28.8%, 23/80), and fully structured (18.8%, 15/80) templates. Automated report dictation clocked in at a median of 180 seconds each, while the other templates took between 210 and 240 seconds.
  • Emergent Connect & Lunit: Emergent Connect and Lunit announced a partnership that will make Lunit’s AI tools available through Emergent Connect’s cloud-based PACS platform. The collaboration extends Lunit’s partner channel (joining Fujifilm PACS and GE hardware), and represents Lunit’s first partnership with a cloud-based PACS developer. The alliance also adds to the list of AI solutions available through Emergent Connect’s PACS platform (joining Zebra and Visionary Health).
  • Self-Referral Delay: CMS will delay its self-referral rule changes until August 2021, after introducing the changes last fall in an effort to streamline compliance for value-based programs. CMS’s planned Stark Law changes could simplify imaging referrals for value-based settings, while still protecting against abuses, but imaging groups have voiced concerns that the changes don’t reflect the way most alternative models actually work (many are still per-service).
  • Bialogics’ NLP Engine: Bialogics launched its new DxPro natural language processing (NLP) engine that integrates into radiology and cardiology PACS workflows to allow providers to mine / analyze their imaging data. Bialogics highlighted DxPro’s potential, noting that NLP can access unstructured data that can’t be extracted using HL7 or DICOM protocols (that’s ~80% of imaging data).
  • MRI CAD for AD: A pair of Turkish scientists developed a MRI CAD system able to distinguish patients with mild cognitive impairment (MCI) due to Alzheimer’s disease (AD) with 87.2% accuracy, representing a step towards developing a non-invasive AD prediction and monitoring system. They developed the algorithm using MRI data from 294 MCI patients captured at the start of the study and again after 12 months (125 developed AD over the follow up period).
  • Ambra & CureMetrix Alliance: Ambra Health and CureMetrix announced a partnership that will make CureMetrix’s AI-based breast cancer screening solutions (cmTriage & cmAssist) available through Ambra Health’s medical image management suite. The alliance advances both companies’ platform-based strategies, as Ambra Health continues to add new partners (also includes AI players like ImageBiopsy Lab and RAPID) and CureMetrix already has platform partnerships with Nuance, Nanox, and Terarecon.
  • Brain Disorder Maps: Georgia State University researchers are using AI to map brain disorder patterns with imaging data from a range of modalities. By combining the imaging data (e.g. brain structure, function, connections) using deep learning algorithms, the researchers plan to develop a framework to help them understand which modalities or brain regions are most relevant to specific disorders and eventually develop multi-modal biomarkers for mental health diagnosis.
  • Hospital Margins Down: Kaufman Hall revealed that hospital operating margins dropped by 96% in the January-July 2020 period (vs. 2019) due to the COVID pandemic, although CARES act funding technically reduced the decline to 28%. During the period, adjusted discharges fell by 13%, adjusted patient days declined by 11%, and emergency department volumes fell by 17%.
  • Fever PET/CT: A study from Johns Hopkins University found that PET/CT is so effective for diagnosing fever of unknown origin (FUO), it deserves wider adoption / coverage, despite its invasiveness and cost downsides. The research team reviewed four studies, finding that FUO PET/CT achieved 97.9% overall sensitivity and a 78.9% overall agreement with other diagnostic techniques.
  • MAGIC’s PICC Impact: The Michigan Appropriateness Guide for Intravenous Catheters (MAGIC) significantly reduced peripherally inserted central catheter (PICC) placements. That’s from a University of Pittsburgh Medical Center study that found that PICC placements by physicians and APPs steadily declined from 243,837 in 2010 to 130,361 in 2018 (46.5%), with the steepest decline corresponding to MAGIC’s 2015 publication.
  • Medicare’s Missing Records: The Office of the Inspector General urged CMS to do a better job collecting unique identifiers from providers who order imaging and other services to prevent fraud. The OIG’s study uncovered that 58% of surveyed Medicare Advantage imaging records from 2018 were missing national provider identifiers, despite 98% of surveyed MA organizations reporting the ability to capture this data. OIG recommends that CMS reject any records with an invalid or inactive identifier.
  • PET/MRI for Lesion Detection: PET/MRI increases lesion detection for some cancers by 15.4% compared to PET/CT, while allowing faster staging and reducing radiation exposure. The German study included 1,003 oncologic examinations from 918 patients who underwent PET/CT and PET/MRI scans. PET/MRI yielded more information in 26.3% of patients, leading to additional malignant findings in 5.3% of patients, and a change in TNM staging in 2.9% of patients.

The Resource Wire

– This is sponsored content.

  • Check out Riverain Technologies’ on-demand webinar demonstrating how its AI solutions integrated into LucidHealth’s radiology workflow and sharing best practices on how to combine AI with radiologist expertise.
  • In this GE Healthcare video, ultrasound users and educators discuss how the Vscan Extend handheld ultrasound combines portability and intuitive design so you can use it in the moment to potentially change patient outcomes.
  • Trying to figure out “Where do we go from here?” The first episode in Bayer Radiology and the AHRA’s COVID-19 rebound podcast series discusses workflow strategies as imaging centers reopen.
  • Did you know that one in three Americans is obese and obesity is even more prevalent in rural communities? This Hitachi blog shows how its wide aperture CT and MRI systems are the best fit for rural hospitals, helping them care for patients of all sizes and get more ROI from their imaging systems.
  • Learn how and why Seattle Children’s Hospital, Duke University Health System, and HCA Healthcare chose to ditch the disk by adopting Nuance’s PowerShare Network.
  • Patients have become savvy healthcare shoppers who increasingly rely on price information to make decisions about their care. Join Healthcare Administrative Partners’ CRO, Rebecca Farrington, as she discusses price transparency & consumerism in radiology in this upcoming RBMA webinar.

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!