Tag: CT scans

Is AI a Help or Hindrance to Radiologists? It’s Down to the Doctor

New research shows AI isn’t always a help for radiologists

Photo by Anna Shvets

One of the most touted promises of medical artificial intelligence tools is their ability to augment human clinicians’ performance by helping them interpret images such as X-rays and CT scans with greater precision to make more accurate diagnoses.

But the benefits of using AI tools on image interpretation appear to vary from clinician to clinician, according to new research led by investigators at Harvard Medical School, working with colleagues at MIT and Stanford.

The study findings suggest that individual clinician differences shape the interaction between human and machine in critical ways that researchers do not yet fully understand. The analysis, published in Nature Medicine, is based on data from an earlier working paper by the same research group released by the National Bureau of Economic Research.

In some instances, the research showed, use of AI can interfere with a radiologist’s performance and interfere with the accuracy of their interpretation.

“We find that different radiologists, indeed, react differently to AI assistance – some are helped while others are hurt by it,” said co-senior author Pranav Rajpurkar, assistant professor of biomedical informatics in the Blavatnik Institute at HMS.

“What this means is that we should not look at radiologists as a uniform population and consider just the ‘average’ effect of AI on their performance,” he said. “To maximize benefits and minimize harm, we need to personalize assistive AI systems.”

The findings underscore the importance of carefully calibrated implementation of AI into clinical practice, but they should in no way discourage the adoption of AI in radiologists’ offices and clinics, the researchers said.

Instead, the results should signal the need to better understand how humans and AI interact and to design carefully calibrated approaches that boost human performance rather than hurt it.

“Clinicians have different levels of expertise, experience, and decision-making styles, so ensuring that AI reflects this diversity is critical for targeted implementation,” said Feiyang “Kathy” Yu, who conducted the work while at the Rajpurkar lab with co-first author on the paper with Alex Moehring at the MIT Sloan School of Management.

“Individual factors and variation would be key in ensuring that AI advances rather than interferes with performance and, ultimately, with diagnosis,” Yu said.

AI tools affected different radiologists differently

While previous research has shown that AI assistants can, indeed, boost radiologists’ diagnostic performance, these studies have looked at radiologists as a whole without accounting for variability from radiologist to radiologist.

In contrast, the new study looks at how individual clinician factors – area of specialty, years of practice, prior use of AI tools – come into play in human-AI collaboration.

The researchers examined how AI tools affected the performance of 140 radiologists on 15 X-ray diagnostic tasks – how reliably the radiologists were able to spot telltale features on an image and make an accurate diagnosis. The analysis involved 324 patient cases with 15 pathologies: abnormal conditions captured on X-rays of the chest.

To determine how AI affected doctors’ ability to spot and correctly identify problems, the researchers used advanced computational methods that captured the magnitude of change in performance when using AI and when not using it.

The effect of AI assistance was inconsistent and varied across radiologists, with the performance of some radiologists improving with AI and worsening in others.

AI tools influenced human performance unpredictably

AI’s effects on human radiologists’ performance varied in often surprising ways.

For instance, contrary to what the researchers expected, factors such how many years of experience a radiologist had, whether they specialised in thoracic, or chest, radiology, and whether they’d used AI readers before, did not reliably predict how an AI tool would affect a doctor’s performance.

Another finding that challenged the prevailing wisdom: Clinicians who had low performance at baseline did not benefit consistently from AI assistance. Some benefited more, some less, and some none at all. Overall, however, lower-performing radiologists at baseline had lower performance with or without AI. The same was true among radiologists who performed better at baseline. They performed consistently well, overall, with or without AI.

Then came a not-so-surprising finding: More accurate AI tools boosted radiologists’ performance, while poorly performing AI tools diminished the diagnostic accuracy of human clinicians.

While the analysis was not done in a way that allowed researchers to determine why this happened, the finding points to the importance of testing and validating AI tool performance before clinical deployment, the researchers said. Such pre-testing could ensure that inferior AI doesn’t interfere with human clinicians’ performance and, therefore, patient care.

What do these findings mean for the future of AI in the clinic?

The researchers cautioned that their findings do not provide an explanation for why and how AI tools seem to affect performance across human clinicians differently, but note that understanding why would be critical to ensuring that AI radiology tools augment human performance rather than hurt it.

To that end, the team noted, AI developers should work with physicians who use their tools to understand and define the precise factors that come into play in the human-AI interaction.

And, the researchers added, the radiologist-AI interaction should be tested in experimental settings that mimic real-world scenarios and reflect the actual patient population for which the tools are designed.

Apart from improving the accuracy of the AI tools, it’s also important to train radiologists to detect inaccurate AI predictions and to question an AI tool’s diagnostic call, the research team said. To achieve that, AI developers should ensure that they design AI models that can “explain” their decisions.

“Our research reveals the nuanced and complex nature of machine-human interaction,” said study co-senior author Nikhil Agarwal, professor of economics at MIT. “It highlights the need to understand the multitude of factors involved in this interplay and how they influence the ultimate diagnosis and care of patients.”

Source: Harvard Medical School

AI-based CT Scans of the Brain can Nearly Match MRI

Photo by Mart Production on Pexels

A new artificial intelligence (AI)-based method can provide as much information on subtle neurodegenerative changes in the brain captured by computed tomography (CT) as compared to magnetic resonance imaging (MRI). The method, reported in the journal Alzheimer’s & Dementia, could enhance diagnostic support, particularly in primary care, for conditions such as dementia and other brain disorders.

Compared to MRI, which requires powerful superconducting magnetics and their associated cryogenic cooling, computed tomography (CT) is a relatively inexpensive and widely available imaging technology. CT is considered inferior to MRI when it comes to reproducing subtle structural changes in the brain or flow changes in the ventricular system. Certain imaging must therefore currently be carried out by specialist departments at larger hospitals equipped with MRI.

AI trained on MRI images

Created with deep learning, a form of AI, the software has been trained to transfer interpretations from MRI images to CT images of the same brains. The new software can provide diagnostic support for radiologists and other professionals who interpret CT images.

“Our method generates diagnostically useful data from routine CT scans that, in some cases, is as good as an MRI scan performed in specialist healthcare,” says Michael Schöll, a professor at Sahlgrenska Academy who led the work involved in the study, carried out in collaboration with researchers at Karolinska Institutet, the National University of Singapore, and Lund University

“The point is that this simple, quick method can provide much more information from examinations that are already carried out on a routine basis within primary care, but also in certain specialist healthcare investigations. In its initial stage, the method can support dementia diagnosis, however, it is also likely to have other applications within neuroradiology.”

Reliable decision-making support

This is a well-validated clinical application of AI-based algorithms, and has the potential to become a fast and reliable form of decision-making support that effectively reduces the number of false negatives. The researchers believe that this solution can improve diagnostics in primary care, optimising patient flow to specialist care.

“This is a major step forward for imaging diagnosis,” says Meera Srikrishna, a postdoctor at the University of Gothenburg and lead author of the study.

“It is now possible to measure the size of different structures or regions of the brain in a similar way to advanced analysis of MRI images. The software makes it possible to segment the brain’s constituent parts in the image and to measure its volume, even though the image quality is not as high with CT.”

Applications for other brain diseases

The software was trained on images of 1117 people, all of whom underwent both CT and MRI imaging. The current study mainly involved healthy older individuals and patients with various forms of dementia. Another application that the team is now investigating is for normal pressure hydrocephalus (NPH).

With NPH, the team has obtained new results indicating that the method can be used both during diagnosis and to monitor the effects of treatment. NPH is a condition that occurs particularly in older people, whereby fluid builds up in the cerebral ventricular system and results in neurological symptoms. About two percent of all people over the age of 65 are affected. Because diagnosis can be complicated and the condition risks being confused with other diseases, many cases are likely to be missed.

“NPH is difficult to diagnose, and it can also be hard to safely evaluate the effect of shunt surgery to drain the fluid in the brain,” continues Michael. “We therefore believe that our method can make a big difference when caring for these patients.”

The software has been developed over the course of several years, and development is now continuing in cooperation with clinics in Sweden, the UK, and the US together with a company, which is a requirement for the innovation to be approved and transferred to healthcare.

Source: University of Gothenburg

MRI-guided Radiation Therapy Reduces Side Effects from Prostate Cancer Radiotherapy

A technique that uses MRI as a guide can make radiotherapy safer for prostate cancer patients by better aiming beams at the prostate while sparing nearby tissue in the bladder, urethra, and rectum. That is the finding of a thorough analysis of all published clinical trials of the technique, called magnetic resonance–guided daily adaptive stereotactic body radiotherapy (MRg-A-SBRT). The analysis is published in CANCER.

By providing detailed images, MRg-A-SBRT can be used to adjust a patient’s radiation plan every day to account for anatomical changes and to monitor the position of the prostate in real time while the radiation beam is on to ensure that treatment is being directed accurately to the prostate. Although MRg-A-SBRT is becoming more popular and multiple clinical trials have tested it, it is unclear whether the technique, which requires more time and resources than standard procedures, has an impact on clinical outcomes and side effects compared with other ways of delivering radiation.

To investigate, Jonathan E. Leeman, MD, of the Dana-Farber Cancer Institute and Brigham and Women’s Hospital, and his colleagues combined data from 29 clinical trials testing MRg-A-SBRT versus conventional CT-guided treatment, with a total of 2457 patients.

MRg-A-SBRT was associated with significantly fewer urinary and bowel side effects in the short term following radiation. Specifically, there was a 44% reduction in urinary side effects and a 60% reduction in bowel side effects.

“The study is the first to directly evaluate the benefits of MR-guided adaptive prostate radiation in comparison to another more standard and conventional form of radiation, and it provides support for use of this treatment in the management of prostate cancer,” said Dr Leeman.

Dr Leeman noted that the study also raises further questions regarding this type of treatment. For example, will the short-term benefits lead to long-term benefits, which are more impactful for patients? Longer follow-up will help answer this question because MRg-A-SBRT is a relatively new treatment. Also, which aspect of the technology is responsible for the improved outcomes seen in clinical trials? “It could potentially be the capability for imaging-based monitoring during the treatment or it could be related to the adaptive planning component. Further studies will be needed to disentangle this,” said Dr Leeman.

An accompanying editorial discusses the analysis’ findings, weighs the potential benefits and shortcomings of adopting this treatment strategy for patients, and questions the value of broad adoption.

Source: Wiley

A Quick Scan Can Pinpoint Hypertension-causing Adrenal Nodules

Stethoscope
Photo by Hush Naidoo on Unsplash

Doctors have demonstrated a new type of CT scan that lights up tiny nodules in the adrenal glands which give rise to hypertension in about 5% of hypertensive patients. enabling hypertension to be cured by their removal. The nodules are discovered in about 5% of hypertensive patients.

Published in The Journal of Hypertension, this work solves a 60-year problem of how to detect the hormone-producing nodules without a difficult and failure-prone catheter study that is available in only a few hospitals. The research also found that, when combined with a urine test, the scan detects a group of patients who come off all their blood pressure medicines after treatment.

The study, led by doctors at Queen Mary University of London and Barts Hospital, and Cambridge University Hospital, involved 128 participants for whom hypertension was found to be caused by aldosterone. The scan found that in two thirds of patients with elevated aldosterone secretion, this is coming from a benign nodule in just one of the adrenal glands, which can then be safely removed. The scan uses a very short-acting dose of metomidate, a radioactive dye that sticks only to the aldosterone-producing nodule.

The scan was as accurate as the old catheter test, but quick, painless and technically successful in every patient. Until now, the catheter test was unable to predict which patients would be completely cured of hypertension by surgical removal of the gland. By contrast, the combination of a ‘hot nodule’ on the scan and urine steroid test detected 18 of the 24 patients who achieved a normal blood pressure off all their drugs.

Professor Morris Brown, co-senior author of the study and Professor of Endocrine Hypertension at Queen Mary University of London, said: “These aldosterone-producing nodules are very small and easily overlooked on a regular CT scan. When they glow for a few minutes after our injection, they are revealed as the obvious cause of hypertension, which can often then be cured. Until now, 99% are never diagnosed because of the difficulty and unavailability of tests. Hopefully this is about to change.”

In most people with hypertension, the cause is unknown, and the condition requires life-long treatment by drugs. Previous research by the group at Queen Mary University discovered that in 5–10% of people with hypertension the cause is a gene mutation in the adrenal glands, which results in excessive amounts of the steroid hormone, aldosterone, being produced. Aldosterone causes salt retention, driving up blood pressure. Patients with excessive aldosterone levels in the blood are resistant to treatment with standard antihypertensives, and at increased risk of cardiovascular disease.

Source: Queen Mary University of London

AI Picks up Incidental Pulmonary Embolism on Chest CT

Photo by Kanasi on Unsplash

According to a study published in the American Journal of Roentgenology, an AI tool for detection of incidental pulmonary embolus (iPE) on conventional contrast-enhanced chest CT examinations had high false negative and moderate false positive rates for detection, and was even able to pick up some iPEs missed by radiologists.

“Potential applications of the AI tool include serving as a second reader to help detect additional iPEs or as a worklist triage tool to allow earlier iPE detection and intervention,” wrote lead investigator Kiran Batra from the University of Texas Southwestern Medical Center in Dallas. “Various explanations of misclassifications by the AI tool (both false positives and false negatives) were identified, to provide targets for model improvement.”

Batra and colleagues’ retrospective study included 2,555 patients (1,340 women, 1,215 men; mean age, 53.6 years) who underwent 3,003 conventional contrast-enhanced chest CT examinations between September 2019 and February 2020 at Parkland Health in Dallas, TX. Using an FDA-approved, commercially available AI tool (Aidoc) to detect acute iPE on the images, a vendor-supplied natural language processing algorithm was then applied to the clinical reports to identify examinations interpreted as positive for iPE.

Ultimately, the commercial AI tool had NPV of 99.8% and PPV of 86.7% for detection of iPE on conventional contrast-enhanced chest CT examinations (ie, not using CT pulmonary angiography protocols). Of 40 iPEs present in the team’s study sample, 7 were detected only by the clinical reports, and 4 were detected only by AI.

Noting that both the AI tool and clinical reports detected iPEs missed by the other method, “the diagnostic performance of the AI tool did not show significant variation across study subgroups,” the authors added.

Source: American Roentgen Ray Society

Amid Shortage, Suggested Ways to Conserve Contrast Agent

Technician and patient with MRI machine
Source: Mart Production on Pexels

Amid an ongoing worldwide shortage of contrast agent for medical imaging, a new UC San Francisco research letter in JAMA described strategies that can be used to safely reduce contrast agent use in computed tomography (CT) by up to 83%.

The three conservation strategies are weight-based (rather than fixed) dosing, reducing contrast dose while reducing tube voltage on scanners, and replacing contrast-enhanced CT with nonenhanced CT when it will minimally affect diagnostic accuracy.

That third strategy – not using the contrast agent in certain CT scans where there is only a small improvement in accuracy – yielded the most dramatic reduction of contrast agent use: 78%.

“Contrast is essential in any situation where we need to assess the blood vessels – for example, for some trauma patients or those with a suspected acute gastrointestinal bleed – and it is also needed for evaluation of certain cancers, such as in the liver or pancreas,” said senior study author Rebecca Smith-Bindman, MD, professor at UCSF.

“However, most CT scans are done for less specific indications such as abdominal pain in a patient with suspected appendicitis,” Prof Smith-Bindman added. “These can and should be done without contrast during the shortage, because the loss of information in these patients will be acceptable for most patients.”

The global shortage of contrast agent started in April with a COVID-related supply chain disruption of GE Healthcare in Shanghai and is expected to last at least several more weeks. More than 54 million diagnostic imaging exams using contrast agents are done every year in the US, a majority being CT scans, and these conservation methods could continue past the current shortage to reduce the use of contrast agent in general, the authors noted.

Referring clinicians are key to conservation
Researchers modelled the three strategies individually and in combination using a sample of 1.04 million CT exams in the UCSF International CT Dose Registry from January 2015 to March 2021.

On its own, weight-based dosing for abdomen, chest, cardiac, spine and extremity imaging reduced contrast agent use by 10%; reducing the tube voltage in appropriate patients allowed a contrast agent reduction of 25%. These two measures combined with using non-contrast CT when possible led to a total reduction of 83%.

Following all three strategies at once may not be possible for some facilities, but each can help conserve supply, Prof Smith-Bindman said. And it is not just radiologists who need to know about them.

“Given the acute shortage, it’s important that clinicians who order imaging exams coordinate with radiology to cancel scans that aren’t absolutely necessary, postpone exams that can be safely delayed, replace CT with MRI and ultrasound where possible, and order an unenhanced scan where possible. Further, clinicians should communicate with their patients about why this is necessary. It is crucial that contrast be conserved for clinical situations where its use is essential for accurate diagnosis,” said Prof Smith-Bindman.

After the shortage ends, medical facilities should consider continuing some of these practices that conserve contrast agent, she added. For example, reducing the tube voltage not only reduces the contrast agent used but also lowers the radiation dose. Tailoring doses weight allows lower dosing volumes for many patients.

In addition, Prof Smith-Bindman noted that this analysis highlights the large amount of contrast agent that is wasted when single-dose vials are used Hospitals and imaging centres that routinely use single-dose contrast agent vials should consider using larger multi-dose vials, which allows for exact dosing and obviates the need to discard unused portions, she said.

“By carrying some of these practices forward, we can mitigate future supply-chain risk and reduce overall waste,” said Smith-Bindman.

Source: University of California – San Francisco

CT Scans Reveal Lung Destruction in Asthmatics

Source: Pixabay CC0

A new study using CT scans have revealed significant changes indicating lung destruction in some asthmatics.

Clinicians have long thought that some people with asthma experience declines in their lung function, called fixed airflow obstruction (FAO), due to changes to their airways. In this study published in The Journal of Allergy and Clinical Immunologyresearchers have found that this issue could extend to the surrounding lung tissue.

Respirologist Kaoruko Shimizu at Hokkaido University said, “Bronchial asthma is considered to be mainly due to inflammation and remodeling of the larger respiratory airways. But not all asthmatics improve with the typical treatments prescribed to alleviate this condition. We wanted to know if changes to the surrounding lung tissue induced a decline in pulmonary function over time in this subgroup of patients.”

Shimizu and her colleagues applied a novel computer tomography (CT)-based approach to detect changes in lung tissue. In this approach, the scientists examined CT scans employing an index called “exponent D” for areas of reduced lung density with increasing coalescence of neighbouring airspaces, which indicates emphysema, or the destruction of air sacs. Airway obstruction was measured by testing the ability of people with asthma to forcefully exhale air in one second. This ability is reduced when the airways are narrower.

The tests examined around 200 smokers and non-smokers with varying degrees of asthma, who were then followed up annually for five years.

People with asthma who experienced persistent airflow limitation, regardless of the severity of their asthma or their smoking status, exhibited constricted airways and also showed signs of lung tissue destruction, the researchers found.

The observed changes to lung tissue in this subgroup of asthmatic patients were not associated with the typical inflammatory markers linked to bronchial asthma. This is important, because it could explain why conventional anti-inflammatory treatments are not as successful in this group.

Future studies should investigate lung destruction in asthma, enabling more personalised management, said Shimizu.

Source: Hokkaido University

CT Scans Improve Outcomes for Concussion Patients

Photo by MART PRODUCTION from Pexels

A study found that CT scans for concussion patients provide crucial information on their risk for long-term impairment and their potential to make a complete recovery, and points to the need for more follow-up.

In the UC San Francisco-led study, researchers examined CT scans of 1935 patients, aged 17 and over, whose neurological exams met criteria for concussion, or mild traumatic brain injury (TBI). Outcomes for moderate and severe TBI have been linked to CT imaging features, but this may be the first time this link has been identified in patients with concussion. This contradicts previous research which had found no prognostic significance of specific types of CT abnormalities.

“Radiologists who routinely read trauma scans know intuitively that patterns of intracranial injury on CT are not random,” said first author Esther Yuh, MD, PhD, of the UCSF Department of Radiology and Biomedical Imaging. “We showed there are patterns of injury, that some of these are associated with worse outcome than others, and that they provide a window into mechanisms of injury that is reproducible across large studies.”

The study was published online in JAMA Neurology.

“Although concussions are referred to as mild traumatic brain injuries, there is nothing mild about some concussions,” explained senior author Geoffrey Manley, MD, PhD, professor and vice chair of neurological surgery at UCSF and chief of neurosurgery at Zuckerberg San Francisco General Hospital. “Patients with concussion may suffer from prolonged headache, poor sleep and impaired concentration, and they are at higher risk of self-medicating with drugs and alcohol. Concussion can also contribute to depression and anxiety, and increase the risk for suicide. We need to view concussion not as an event but as a disease requiring physician follow-up after a patient is discharged from the hospital.”

The participants were enrolled by the brain injury research initiative TRACK-TBI, of which Manley is the principal investigator. To enrich the number of so-called complicated concussions, the researchers drew exclusively from patients who had been seen at hospitals with level 1 trauma centres. This meant 37 percent of study participants had a positive CT, significantly more than the 9 percent of positive CTs from patients in US emergency departments.

The most common patterns of injury, affecting more than half of CT-positive patients, were combinations of subarachnoid haemorrhage (SAH), subdural haematoma (SDH), and/or contusion, which may be caused by injuries such as falls from standing. About 7 percent had intraventricular haemorrhage (IVH) or petechial haemorrhage, caused by head rotation as in some sporting, scooter and automobile accidents; and 5 percent had epidural haematoma (EDH), often seen in sports injuries such as being hit with a baseball.

Average age of the patients was 41 and 66 percent were male. They were followed-up at two weeks, and at three-, six- and 12 months following injury. Patients in the SAH/SDH/contusion group failed to make a complete recovery at 12 months post-injury and had a range of outcome impairments, from mild to more severe.

Patients in the IVH/petechial haemorrhage group tended toward more severe impairments, in the lower-moderate disability range, a level potentially affecting multiple areas of function, such as employment, social and leisure activities, up to 12 months post-injury. Patients with EDH fared significantly better and demonstrated complete recovery by their six-month assessment.

Results from CENTER-TBI, a parallel brain injury research group that had enrolled 2594 participants at European trauma centres. validated the findings. “The confirmation of the findings in an independent cohort confirms the fidelity of our results,” said Manley, adding that patients with EDH were one exception, with incomplete recovery lingering for months longer than those patients followed by TRACK-TBI. However, more severe outcomes were not seen at any point in either study.

The researchers noted that even among concussion patients with positive CT scans, only 39 percent get follow-up care, which should be routine. They also cautioned that their findings are not a call for increased CT use, which has radiation dose concerns and is restricted to known or suspected concussions.

Indeed, a recently approved rapid hand-held blood test may reduce the amount of CT scans. Manley found this test was more sensitive than CT in detecting concussion. The blood test measures biomarkers associated with TBI, which were nearly 52 times higher in MRI-identified concussion patients than in healthy participants.

In addition to challenging the belief that CT features in concussion are not relevant, the researchers are also challenging the idea that concussion is “what the patient brings to the injury,” said Manley, who is also affiliated with the UCSF Weill Institute for Neurosciences. “In moderate and severe TBI, it is anecdotally taught that outcome is determined by ‘what the injury brings to the patient,’ while concussion is determined by baseline characteristics like age, sex and years of education. While the study confirms the importance of these characteristics, we show that in some concussion cases, poor outcomes are also attributed to ‘what the injury brings to the patient.'”

Source: University of California, San Francisco

Journal information: Yuh EL et al., Pathological computed tomography features associated with adverse outcomes after mild traumatic brain injury, JAMA Neurology, July 19, 2021.