Uncategorized

Military AI | Selfie Screening | Triage Works

“. . . this initiative has the potential to compliment other significant military medical advancements to include antisepsis, blood transfusions, and vaccines.”

U.S. DoD Joint Artificial Intelligence Center chief, Dr. Hassan Tetteh, sharing his big goals for the U.S. military’s new imaging AI initiative.


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


Military AI

The U.S. Department of Defense revealed plans to develop a portfolio of medical imaging AI solutions, combining its massive / diverse imaging datasets with “the best of commercially available artificial intelligence technology.” This story hasn’t received much coverage, but this is technically one of the world’s largest and most technologically advanced healthcare systems making a push into imaging AI. That’s pretty notable.

  • AI Mission – The DoD’s Predictive Health project will develop a portfolio of cancer-detecting AI tools over the next 24 months. The tools could play a key role in the Joint Artificial Intelligence Center’s (JAIC) “Warfighter Health Mission,” which recently launched with the goal of transforming military healthcare through AI.
  • The AI Team – In addition to the Joint Artificial Intelligence Center, the Predictive Health initiative will involve the DoD’s Defense Innovation Unit, the Defense Health Agency, and three unnamed private companies.
  • The AI Tools – The new AI tools will focus on early cancer detection, using both radiographic and pathologic images.
  • Four Priorities – The DoD’s imaging AI mission has four main objectives: 1) Reducing healthcare costs through early detection; 2) “Maximizing personnel readiness” by minimizing its sick service members; 3) Streamlining military clinician workflows; and 4) Optimizing soldier performance and saving lives by catching cancer earlier and avoiding invasive procedures.
  • Why this is Big – Given the military’s unique resources (huge & diverse datasets, advanced tech, tons of funding), this initiative could bring major advancements in imaging AI science and adoption. It even has the potential to be as big as the military’s previous breakthroughs with “antisepsis, blood transfusions, and vaccines.”



AI Triage Works

A new study out of Sweden found that using AI-based breast cancer detection software to triage and remove “clearly negative” mammograms could significantly reduce radiologists’ reading volumes without missing cancers. It also could reduce false negatives and allow earlier detection.

  • The Triage Study – The researchers used screening mammograms from 7,364 women (547 w/ breast cancer) and a commercially available AI tool to create cancer prediction scores. They then evaluated cancer rates at different AI score levels, calculating the percent of missed cancers if the women from different score levels were screened using an AI-only workflow (no radiologists).
  • The Triage Results – The study found that none of the women with the lowest 60% of AI scores would have missed cancer diagnoses if radiologists never reviewed their scans (lowest 70% = 0.3% missed cancers, lowest 80% = 2.6% missed).
  • The Enhanced Workflow – The researchers had radiologists review mammograms from the women with the highest 40% of AI scores, and then performed “enhanced assessments” on the women with negative screenings and high risk scores. If the enhanced assessments focused on the highest 1% or 5% of AI scores, it would have detected 24 (12%) or 53 (27%) of the 200 subsequent interval cancers, and 48 (14%) or 121 (35%) of the 347 next-round screen-detected cancers.
  • Conclusion – Noting that only 5 out of 1,000 screenings lead to cancer detections, and another two cancers out of those 1,000 screenings are missed, a process like this could reduce radiologist labor and improve detection among women with the highest risks.

The Wire

  • Selfie Screening: Digital photo ‘selfies’ of patients’ faces processed with an AI algorithm could be used to screen for heart disease. That’s from new research out of China, were scientists developed an algorithm that can detect coronary artery disease (CAD) by analyzing four photographs of a person’s face (n = 5,216 people training, 580 validation, 1,013 testing), achieving 80% sensitivity and 54% specificity in external testing (80% and 61% in validation). This isn’t our type of imaging, but patients flagged due to ‘selfie’ screenings would likely be sent to a CCTA exam.
  • Imaging Cybersecurity: A team from Israel’s Ben-Gurion University unveiled a new AI-based technique intended to protect medical imaging systems from cyberattacks and prevent human and system errors. The new technique analyzes instructions sent from PCs to imaging systems, identifying abnormal instructions (e.g. 100x higher radiation dosage) or spotting instructions that don’t match patient information (e.g. scan type doesn’t match patient’s age, weight, or potential diagnosis).
  • EchoNous AI: EchoNous took a step towards breaking the ultrasound adoption barrier, launching its new ‘Trio’ series of algorithms for use with the company’s Kosmos point-of-care ultrasound system. The AI tools aim to simplify three of ultrasound’s most common operational challenges—imaging guidance, quality grading, and anatomical labeling—to allow more medical students and doctors to use ultrasound at patient bedsides.
  • Different Software, Different Results: A team of Austrian researchers found that different automated MRI perfusion-diffusion software (in this case RAPID and Olea Sphere) could produce “significantly different” results. The team analyzed brain MRIs from 81 patients with acute stroke due to anterior circulation LVO, finding significantly different measurements for hypoperfused brain tissue (median: 91.0 ml RAPID vs. 102.2 ml Olea Sphere), ADC volume (30.0 ml vs. 23.9 ml), and perfusion-diffusion volume (47.0 ml vs. 67.2 ml). These results suggest that a hospital’s choice of software might “seriously influence” their decisions whether to perform mechanical thrombectomy.
  • Mental Health Imaging: Researchers at the University of Tokyo developed a machine learning algorithm to perform psychiatric diagnoses based on brain MRI scans, representing a step toward reducing mental health diagnostic uncertainty. The team applied brain MRIs from 206 Japanese adults to develop an algorithm that can identify the brains of nonpatients, patients diagnosed with autism, and patients with schizophrenia. When tested against scans from 43 additional patients, the algorithm matched psychiatrists’ assessments with 85% accuracy, with cerebral cortex thickness proving to be the most distinguishing feature.
  • See-Mode’s $7M: See-Mode Technologies, a stroke prediction and prevention AI startup out of Singapore and Australia, just completed a $7 million financing round, following a $1m seed round in 2018. See-Mode will use the money to double its business development and R&D headcount.
  • Pre-Surgical AR Benefits: A new study out of UPenn found that using augmented reality (AR) to view MR images before performing transarterial embolization for hepatocellular carcinomas could make these procedures more efficient. The researchers used 3D AR systems to view pre-procedural MR images from 12 rats with HCCs before performing embolization, finding that the AR step reduced fluoroscopy times by 48% (14.1min to 7.4min) and cut catheterization time by 27% (42.7 min to 31 min).
  • Appendicitis CTs & the Cognitively Impaired: A study by the American College of Surgeons found that pediatric laparoscopic appendectomy patients are far more likely to receive CT scans if they have cognitive impairments, due in part to communication difficulties (55% vs. 41% without developmental issues). Their analysis of 16,986 pediatric laparoscopic appendectomy patients also showed that the 293 children with cognitive impairments had higher rates of readmission (8% vs. 3%) and postoperative emergency room visits (13% vs. 8%).
  • Imaging Biometrics on Arterys Platform: Imaging Biometrics’ suite of brain tumor visualization tools are now available on the Arterys Marketplace AI platform, including its IB Neuro (MRI dynamic susceptibility contrast solution), IB DCR (Delta T1 mapping algorithm), and IB Diffusion (analyzes MR diffusion-weighted images) tools. Imaging Biometrics joins a growing range of Arterys and 3rd party solutions available on the Arterys Marketplace.
  • CAC Heart Attack Predictor: Researchers at UT Southwestern found that CAC scores are more effective for predicting heart attacks than strokes. The team analyzed data from 7,042 patients over 12.3 years, finding that patients with coronary artery calcium levels at or above 100 had a twofold greater risk of heart attack than stroke, compared to even heart attack and stroke risks in patients with 0-99 CAC scores.
  • Ultrasound’s Post-DBT Advantage: Ultrasound is the most common modality for performing workups on breast masses identified during DBT screening exams and could replace the use of DM + US workups for these women (thus reducing radiation, cost, and time). This is based on a University of Texas Southwestern study that reviewed 266 non-calcified lesions detected during DBT screenings (247 women), finding that ultrasound was used in 69% of breast mass workups (8x greater odds than DM + US) and identified more true lesions. However, DM+US workups were still superior to US-only workups for architectural distortions and focal asymmetries found in DBT screenings.
  • Unprepared for MTBs: Most radiologists attend multidisciplinary tumor board (MTB) meetings, but less than half review exams beforehand, according to a survey of 292 European Society of Oncologic Imaging (ESOI) members. Only 43.9% of survey respondents reviewed more than 70% of exams before MTBs, with many of them citing time constraints and busy schedules.

The Resource Wire

– This is sponsored content.

  • Wondering what to look for in your next CT? This Hitachi blog goes beyond slice count, detailing the top three features that your next CT has to have. Here’s a hint: they will help you care for your >30% patients who are overweight.
  • Radiology will see a significant cut in Medicare reimbursement in 2021 if the MPFS Proposed Rule is applied without a change to the budget neutrality requirement in the law. Get all the details in this Healthcare Administrative Partners blog post.
  • This Riverain Technologies case study details how Einstein Medical Center adopted ClearRead CT enterprise-wide (all 13 CT scanners) and how the solution allowed Einstein radiologists to identify small nodules faster and more reliably.
  • Severe sepsis strikes more than a million Americans every year at an annual cost of more than $20 billion. Learn how point of care ultrasound can help improve sepsis outcomes in this GE Healthcare paper.
  • Bayer’s new Gadavist Imaging Bulk Package multi-patient dosing system eliminates the waste associated with individual GBCA vials, benefiting rad techs (improved workflow, increased patient focus), administrators (reduced costs / waste, increased productivity), and patients (greater dosage consistency).

You're signed up!

It's great to have you as a reader. Check your inbox for a welcome email.

-- The Imaging Wire team