“Am I the only one getting frustrated at all press on ‘AI is better than Docs’ or ‘will AI take my jobs’?? I thought we are here for the patient. To give him/her better quality care. #AI is just a means to that end. Why do people forget that?”
Qure.ai’s Chiranjiv Singh on the limited role of patient care in many AI articles and editorials.
Imaging Wire Sponsors
- Carestream – Focused on delivering innovation that is life changing – for patients, customers, employees, communities and other stakeholders
- Focused Ultrasound Foundation – Accelerating the development and adoption of focused ultrasound
- Medmo – Helping underinsured Americans save on medical scans by connecting them to imaging providers with unfilled schedule time
- Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter
- Pocus Systems – A new Point of Care Ultrasound startup, combining a team of POCUS veterans with next-generation genuine AI technology to disrupt the industry
- Qure.ai – Making healthcare more accessible by applying deep learning to radiology imaging
The Imaging Wire
There’s been a lot of news of AI catching signs of cancer “x-years” before it’s visible, but a new study in the British Journal of Radiology suggests that human radiologists also can “detect the ‘gist’ of breast cancer in mammograms three years before localized signs of cancer are visible.” Here’s how they figured this out:
- In 4 prospective studies, 59 expert observers from 3 groups viewed 116–200 bilateral mammograms for 500 milliseconds each
- Half of the images were from exams performed three years before onset of visible/actionable cancer, while half were normal images
- The observers rated likelihood of abnormality on a 0–100 scale, with all three groups detecting abnormal images at above chance levels (0.53, 0.54, 0.54), with the most experienced radiologists achieving the greatest accuracy
This is a good sign for early breast cancer detection, but much like AI’s black box, the greatest breakthroughs could come from understanding what goes on in radiologists’ minds when they “gestalt” an image or a study like this. This understanding might actually make “gist”-based diagnoses actionable (rather than a huge over-diagnosis risk), lead to future hardware and software innovations, and help the radiology community better understand how their human instincts would be best complimented by AI. At least until we train AI to beat radiologists’ subconscious.
BU Turns up the MRI Volume
A team of Boston University scientists developed a new metamaterial that can increase the imaging power of lower-strength MRIs and cut scan times in half, potentially having a major impact on hospitals and imaging centers (operational efficiency, equipment costs, magnetic field risks) and helping expand MRIs to developing countries.
- Low to High – The BU team is specifically targeting their new magnetic metamaterial for use with ultra–low field MRI, where its image clarity and speed improvements would have the greatest impact. When tested on a 1.5T MRI scanning chicken legs and fruits, the metamaterial improved the MRI’s signal-to-noise ratio by 4.2-times, suggesting that lower magnetic fields could be used to produce clearer images than currently possible (on actual humans).
- Simple for Complex – Although MRI may be “one of the most complex systems invented by human beings,” the metamaterial is relatively simple. It’s built from flexible arrays of helical resonators (3cm structures made from 3D-printed plastic and coils of thin copper wire) that interact with MRI’s magnetic field and boost its signal-to-noise ratio, thus “turning up the volume of the image.”
Next up, the BU researchers will seek industry partners to help drive adoption of the magnetic metamaterial for clinical applications.
AI for IR
Although AI remains almost completely associated with diagnostic radiology, a new editorial in the American Journal of Roentgenology argues that machine learning could have a significant impact on interventional radiology (IR) in the future. The paper suggests that ML’s ability to support image analysis is likely to provide value in the angiography suite and its value could extend to “patient selection modeling, predictive tools for treatment planning, trainee education, and others.”
- Imaging Analysis – Despite IR’s slow start with AI/ML-based imaging analysis, this may change as IR researchers take note of how artificial intelligence is supporting diagnostic radiology, noting that intraprocedural IR angiographic analytics could adopt algorithms similar to the FFR (among other adaptations of diagnostic AI/ML solutions).
- Treatment Modeling – ML may be able to fuel clinical decision support (CDS) systems, thus supporting IR’s treatment decisions.
- Education – The combination of ML and augmented reality could prove to be a valuable method to give IR trainees experience simulating certain procedures or to assess trainees’ capabilities.
- A new report finds that the UK NHS is experiencing significant ultrasound sonographer shortages, as a combination of rising ultrasound procedure volumes (up 2.5x since 1996), departing personnel, and insufficient applicants are bringing NHS ultrasound departments “very close to breaking point.” Although the shortage is partially influenced by retirements, one major challenge is that trained sonographers are leaving for “better positions,” forcing departments to use higher-cost temporary sonographers and undermining training initiatives.
- Aidoc announced the FDA clearance of its AI solution to triage cervical spine fractures, which automatically prioritizes suspected C-spine fracture cases in a radiologist’s worklist to ensure the most critical cases are diagnosed first. This is Aidoc’s third FDA clearance in the last 9 months (PE in May, ICH in August 2018), which is pretty impressive, and attributed to the company’s scalable AI engine.
- Routine breast cancer screening guidelines continue to be a topic of significant debate, but an Oklahoma State University study at least confirmed that physicians who helped develop the various consensus breast cancer screening recommendations were not influenced by financial conflicts of interest (e.g. payments from drug and device companies). The study looked at 43 authors from 7 guideline documents, finding that the majority did not receive an industry payment (29 no payment, 14 at least one). Although five of the authors received more than $5,000 from a single company in a single year, this was still well below previous research that found 40% of healthcare guideline authors received at least $5,000 in annual payments.
- Researchers in Japan found that deep convolutional neural networks (DCNNs) can be more accurate when trained on augmented datasets, which are created by altering the original images (rotation, blurriness, brightness, etc.) to increase the size of the dataset. The study used an original dataset of 288 abnormal and 447 normal radiographs (441 for training & validation, 294 for testing), and then augmented the training images to create 12,789 augmented images. Algorithms trained on the augmented dataset identified abnormal and normal chest radiographs more accurately (highest sets = 0.91).
- Siemens Healthineers launched a research and education program with Holland’s University of Twente (UT), focused on improving image and robot-guided surgery technology. Through the partnership, Siemens Healthineers is providing UT with a robot-guided 3D imaging system for use in its training operating room as well as an MRI to support research into expanding the modality’s use in minimally-invasive procedures.
- A Japanese research team found that whole-body CT images scanned in ultrafast scan mode in emergency department settings achieved a significant reduction in motion artifacts versus conventional CTs. The team looked at scans from 60 unconscious patients (30 with new generation CT in ultrafast mode, 30 with conventional CTs), finding that ultrafast mode significantly reduced artifacts in the aortic root (p = 0.0003), lower lungs (p = 0.011), diaphragm (p = 0.0047), liver (p = 0.0026), and kidneys (p = 0.019), but not in the aortic arch, thoracic descending aorta, abdominal aorta, and upper lungs.
- Hitachi is reportedly planning to make Hitachi High-Technologies Corp., which specializes in medical equipment and semiconductor production, a fully owned subsidiary (up from its 51.7% stake). Hitachi High-Technologies Corp may not be directly involved in imaging, but this move will grow and diversify Hitachi’s imaging-centric medical business unit, helping the company towards its goal of growing the unit’s revenue to ¥400 billion by fiscal 2021 (vs. ¥360 billion).
- Canon Medical Systems UK will soon provide multi-vendor ultrasound service to its hospital clients through a partnership with independent service provider, Imagex Medical. Canon highlighted how UK hospitals would be able to use the new multi-vendor service to streamline their ultrasound service relationships into one single contract (vs. 2-4 single-brand OEMs visiting at different times), providing operational and cost savings, while Canon also surely aims to use this offering to gain full control over its multi-brand ultrasound accounts.
- Carestream introduced its newly-designed small-format DRX Plus 2530C Detector (225×30, 98 microns), which features lower-dose cesium iodide technology, and is intended for pediatric, extremity, and tabletop imaging. The DRX Plus 2530C launches with a number of key improvements to Carestream’s first small-format detector, adding beam-sensing technology, a built-in wireless access point to simplify installation, and a longer-life battery.
- A team of German researchers found that 68Ga-FAPI PET/CT radiotracers can effectively identify 28 types of malignant tumors with very high uptake and image contrast, potentially allowing new applications for characterization, staging, and therapy. This follow-up to a March study that highlighted 68Ga-FAPI’s effectiveness (in comparison to 18F-FDG radiotracers) used PET/CT images from 80 patients with 28 different kinds of cancer to quantify 68Ga-FAPI uptake, finding that the new radiotracer achieved the highest uptake (SUVmax >12) in sarcoma, esophageal, breast, cholangiocarcinoma, and lung cancer (many of which are weak areas for 18F-FDG traces).
- Chinese imaging AI company, Deepwise, bolstered its already-“deep” pockets with the completion of its latest Series C funding round. Although Deepwise didn’t disclose its latest funding amount, its first three funding events raised an estimated $51 million (including $22.6m in 2018), making it among the most capitalized companies in imaging AI.
- Noting the North America healthcare system’s shift to online parts and supplies procurement and its own goals to expand its customer relationships, GE Healthcare has expanded its inventory of multi-vendor parts available through its online GE Healthcare Service Shop.
The Resource Wire
– This is sponsored content.
- Qure.ai was selected as a top 3 finalist at the ITU AI for Good summit in Geneva, a prestigious United Nations platform that fosters dialogue on the beneficial use of Artificial Intelligence. Here’s Qure.ai’s AI for Good presentation (starts around 25 minute mark).
- In this Carestream video, an Orthopedic surgeon discusses how the OnSite Extremity CT has improved his business and the care he provides his patients.
- Focused Ultrasound Foundation founder and Chairman, Neal F. Kassell, MD, took to the foundation’s blog to emphasize the critical role of industry collaboration in its success and to encourage ongoing collaboration.
- The University of Rochester Medical Center’s adoption of Nuance mPower Clinical Analytics and PowerScribe Follow-up Manager solutions brought significant improvements to its Backstop follow-up tracking program. URMC now satisfactorily closes 91% of its 500 tracked monthly recommendations, reducing the risk of delayed diagnosis by 80% and increasing its examination completion rate by 29% (from 55% to 71%).
- POCUS Systems is approved as a Veteran Owned Business with the US Government Office of Veterans Business Development, paving the way for partnerships with the federal healthcare delivery systems.
- By partnering with Medmo, imaging centers can keep their schedules full, their equipment busy, and increase revenue. Here’s where to get started.