|
Right Diagnoses, Wrong Reasons | RSNA Preview
November 15, 2021
|
|
|
|
|
Together with
|
|
|
|
“You don’t say it’s cortical bone weather.”
|
|
Dr. Glaucomflecken’s latest radiologist therapy session, working on the right way to say that it’s “really bright outside” when radiologists venture out of the reading room.
|
|
|
I’m happy to share the latest Imaging Wire Show featuring Siemens Healthineers’ Kelly Parker. Join us for a deep dive into MR’s accessibility evolution, and the technology and design breakthroughs that are bringing MR into completely new settings.
|
|
|
Photo: American Journal of Roentgenology |
|
|
An AJR study shared new evidence of how X-ray image labels influence deep learning decision making, while revealing one way developers can address this issue.
Confounding History – Although already well known by AI insiders, label and laterality-based AI shortcuts made headlines last year when they were blamed for many COVID algorithms’ poor real-world performance.
The Study – Using 40k images from Stanford’s MURA dataset, the researchers trained three CNNs to detect abnormalities in upper extremity X-rays. They then tested the models for detection accuracy and used a heatmap tool to identify the parts of the images that the CNNs emphasized. As you might expect, labels played a major role in both accuracy and decision making.
- The model trained on complete images (bones & labels) achieved an 0.844 AUC, but based 89% of its decisions on the radiographs’ laterality/labels.
- The model trained without labels or laterality (only bones) detected abnormalities with a higher 0.857 AUC and attributed 91% of its decision to bone features.
- The model trained with only laterality and labels (no bones) still achieved an 0.638 AUC, showing that AI interprets certain labels as a sign of abnormalities.
The Takeaway – Labels are just about as common on X-rays as actual anatomy, and it turns out that they could have an even greater influence on AI decision making. Because of that, the authors urged AI developers to address confounding image features during the curation process (potentially by covering labels) and encouraged AI users to screen CNNs for these issues before clinical deployment.
|
|
|
COPC Ditches the Disk with Novarad
See how Novarad’s CryptoChart solution allowed Central Ohio Primary Care (COPC, 70 practices, 400 physicians) to make the transition to digital imaging sharing in this Healthcare IT News case study.
|
|
HUMC’s Case for Siemens Healthineers TrueV
When the demand for your PET/CT imaging services outpaces available appointments, what are your options? Learn how Hackensack University Medical Center optimized its clinical operations by upgrading its Biograph Horizon to TrueV technology in this new case study from Siemens Healthineers.
|
|
- RSNA Expectations: We’re two weeks away from the first in-person RSNA since COVID and it appears that the folks who are headed to Chicago plan to make it count. RSNA already confirmed that it will be much smaller this year, but most of the expected exhibitors and people are going, and we’re not seeing late cancellations like we did with HIMSS (thanks, Delta). Based on early announcements, it looks like we’ll see some solid modality innovations (low-field MR, Photon-Counting CTs), continued cloud and AI-enabled PACS momentum, and a far more mature lineup of AI exhibitors.
- LDCT Matches CT for Appendicitis: A new study out of Finland found that low-dose CT matches standard CT for appendicitis assessments, suggesting that LDCT should become the standard clinical practice for non-obese patients. The researchers performed LDCT or CT exams on 856 patients with suspected appendicitis (454 w/ LDCT), finding that LDCT matched or surpassed CT for detecting appendicitis (98% vs. 98.5%) and differentiating between uncomplicated and complicated cases (90.3% vs. 87.6%). LDCT also outperformed CT when differentiating complication levels among patients with a BMI under 30 kg/m2 (89.8% vs. 88.4%).
- Imaging’s Positive Q3: Despite supply chain headwinds, most major Imaging OEMs reported solid healthcare/imaging division revenue growth for the July-September quarter. Fujifilm once again posted the greatest healthcare growth due in part to its Hitachi integration (+46% to $1.8b), followed by Konica Minolta’s healthcare division (+24% to $271m), Hologic’s breast imaging division (+12.7% to $265m), Siemens Healthineers’ imaging business (+11.6% to $3.1b), Philips’ Diagnosis & Treatment division (+10% to $2.4b), and Canon Medical Systems (+9.8% to $1b). Meanwhile, supply chain issues caused GE Healthcare’s revenue to fall in Q3 (-5% to $4.3b).
- MSK’s Volume Prophecy: Memorial Sloan Kettering radiologists used Facebook/Meta’s open-source forecasting tool (Prophet) to accurately predict imaging volumes, showing how a tool like this could support resource planning. The team trained the Prophet tool with its daily CT and MRI volume data (610,570 exams, 2015-2019) and used it to accurately predicted MSK’s February 2020 CT (9,934 predicted vs. 9,667 actual) and MRI (2,484 vs. 2,457 actual) exam volumes – far more accurately than MSK’s manual approach. Even after major COVID-driven imaging shifts, the Prophet tool achieved similar accuracy in another August 2020 test.
- NHS Digital Push: The UK government will give the NHS £248m over the next year to digitize its diagnostic workflows, bringing new solutions for image sharing, remote image reading, and appropriate imaging referrals. These upgrades are intended to allow the NHS to overcome its well-publicized diagnostic imaging backlog and help it prepare for a major initiative that will add over 100 community imaging centers during the next three years.
- Siemens & UCSF Go Green: Siemens Healthineers and UCSF announced a new green radiology alliance that will lead to the first carbon-neutral radiology imaging service. UCSF’s carbon-neutral strategy will initially focus on improving patient access to scanners (thus reducing patient commutes), adopting Siemens’ unique MAGNETOM Free.Max MRI (low helium usage, small weight & footprint, flexible installation), and monitoring imaging power consumption.
- Patient Satisfaction Drives Imaging Volumes: Ohio State researchers recently found that patient satisfaction might actually drive MRI scan volumes. Analysis of nine outpatient MRI sites during a one-year period (39,595 patient visits) found that OSU’s sites that increased their MRI volumes by >5% also saw their satisfaction scores increase, indicating that “patient experiences or perceptions of quality” might influence which centers they visit. Some might debate correlation and causation, but given the consumerization of outpatient imaging, this theory makes sense.
- Mirion Acquires CIRS: Mirion Technologies continued its healthcare expansion, acquiring medical imaging and radiation therapy phantoms company, Computerized Imaging Reference Systems (CIRS) for $54m. Mirion has been a major player in the radiation measurement industry for years, and began its expansion into medical imaging when it acquired Biodex Medical Systems just over a year ago.
- CXR ED AI: A new study out of South Korea found that emergency physicians’ CXR interpretations improved when they used Lunit’s INSIGHT CXR AI tool, especially inexperienced physicians (<2yrs experience). The researchers had seven ED physicians review 388 CXR cases to detect abnormalities and make clinical decisions (259 had abnormalities), first only using case information and then using INSIGHT CXR. The AI step led to far more clinical decision changes among the inexperienced physicians than the experienced physicians (106 vs. 20 changes), while AI improved abnormality detection accuracy with both groups (inexperienced: 70.23% to 76.35%; experienced: 71.39% to 76.37%).
- Novarad OpenSight for Education: Novarad launched an education version of its OpenSight Augmented Reality System, allowing students and teachers to interact with CT scans using Microsoft HoloLens 2 headsets, and giving them a 360-degree view of any CT slice in a study. We’re seeing VR gain momentum in both the clinic and the classroom, making this a logical expansion for Novarad.
- Philips’ Gates Grant: Philips received a $15.4m grant from the Bill & Melinda Gates Foundation that will help fund the development of AI-based handheld ultrasound apps intended to improve obstetric care in developing countries. The apps will help front-line healthcare workers (e.g. midwives) acquire the right images using Philips’ Lumify ultrasound and then assist with image interpretation, allowing them to identify more early-stage pregnancy problems and reduce childbirth deaths and fetal mortality.
|
|
Riverain’s Lung Cancer Guideline Kit
The USPSTF lung cancer screening guidelines were updated in May 2021, and driving compliance to these guidelines is a long, slow, repetitive process. Because of that, the Riverain team put together this kit to help hospitals and imaging centers educate their referring physicians or patients on the new guidelines.
|
|
- Check out this Imaging Wire Show interview with Blackford Analysis founder and CEO, Ben Panter, detailing how to solve AI’s assessment and deployment problem, AI’s downstream value, and what it will take for AI to have its greatest impact.
- Why United Imaging’s MI (uMI)? Every United Imaging molecular imaging system features its “uEXPLORER Inside” technology platform, which is designed for total-body scanning, is scalable for clinical systems, and excellent in an MR environment – you’ll see a big difference and your patients can benefit from their focus on coverage, clarity, and sensitivity.
- Workflow improvement, remote access, high image quality, and some major time savings. These are just a few of the benefits that cardiologists and their teams experienced with Fujifilm Healthcare’s VidiStar platform.
- Learn how Einstein Healthcare Network leveraged Nuance’s PowerScribe One platform and the Nuance AI Marketplace to put AI into action.
- The need for faster radiology report turnaround is increasingly clear, but the ways to improve turnaround time aren’t always. See how smarter prioritization, streamlined communication, and true integration are driving faster turnarounds in this new GE Healthcare report.
- This MDPI study details how imaging represents the most powerful tool for identifying patients at risk of arrhythmic mitral valve prolapse, noting Artery’s 4D-flow CMR’s “great promise” for assessing mitral regurgitation.
|
|
|
|
|
