• Our Partners
  • CarePolicy
  • HomeCareConsulting
  • Digit9X
  • Home
  • Assisted Living
  • Elderly
  • Home Care Agency
  • Home Care Worker
  • Home Nursing
Menu
  • Home
  • Assisted Living
  • Elderly
  • Home Care Agency
  • Home Care Worker
  • Home Nursing
Home » Break the “intellectual bottleneck”: How AI calculates things that could not be compartmented previously in healthcare
Home Care Worker

Break the “intellectual bottleneck”: How AI calculates things that could not be compartmented previously in healthcare

adminBy adminApril 30, 2025No Comments7 Mins Read
Facebook Twitter Pinterest Telegram LinkedIn Tumblr Email Reddit
Share
Facebook Twitter LinkedIn Pinterest Email Copy Link


Join our daily and weekly newsletter for the latest updates and exclusive content on industry-leading AI coverage. learn more

Whenever a patient undergoes a CT scan at the University of Texas Department of Medicine (UTMB), the resulting images are automatically sent to the Cardiac Disease Department, analyzed by AI, and assigned a cardiac risk score.

In just a few months, thanks to simple algorithms, AI flagged multiple patients with high cardiovascular risk. A CT scan does not need to be heart-related. The patient does not need to have heart problems. All scans will automatically trigger an evaluation.

This is a simple preventive care enabled by AI, allowing medical facilities to eventually begin to take advantage of the vast amount of data.

“The data is just sitting there,” Peter McCaffrey, head of AI at UTMB, told VentureBeat. “What I like about this is that AI doesn't have to do superhuman things. It's doing low intelligence tasks, but it's a huge amount of stuff and it still offers a lot of value.

He said, “We know we miss things. We just didn't have the tools to go back and find.”

How AI helps UTMB determine cardiovascular risk

Like many healthcare facilities, UTMB applies AI to many areas. One of the first use cases is cardiac risk screening. The model is trained to scan for accidental coronary calcification (ICAC), a powerful predictor of cardiovascular risk. The goal is to identify patients susceptible to heart disease, which may be overlooked because they do not show obvious symptoms, McCaffrey explained.

Through the screening program, all CT scans completed at the facility are automatically analyzed using AI to detect coronary calcification. Scanning has nothing to do with heart disease. It may be ordered due to spinal fractures or abnormal pulmonary nodules.

The scan is fed to an image-based convolutional neural network (CNN) that calculates an Agathstone score representing the accumulation of plaque in the patient's artery. McCaffrey explained that this is usually calculated by a human radiologist.

From there, AI assigns patients with an ICAC score of 100 or more to three “risk layers” based on additional information (either using statins or visiting with a cardiologist). McCaffrey explains that this assignment is rule-based and can be derived from discrete values ​​in electronic health records (EHR), and AI can determine the values ​​by using GPT-4o to process free texts such as clinical visit notes.

Patients flagged with a score of 100 or more without a known history of heart disease visits or treatment will automatically be sent a digital message. The system also sends notes to the major doctors. Patients identified as having a more serious ICAC score of 300 or higher will also receive calls.

McCaffrey explained that almost everything is automated, except for the phone. However, the facility is a tool that actively pilots the game in the hopes of automating voice calls as well. The only area where humans are in the loop is to check the AI-derived calcium score and risk layer before proceeding with automatic notifications.

Since launching the program in late 2024, health facilities have evaluated approximately 450 scans per month, with five to 10 of these cases being identified as high-risk each month, and intervention is required, McCaffrey reports.

“The point here is that no one suspects you have this disease, and no one needs to order a study of this disease,” he pointed out.

Another important use case for AI is the detection of stroke and pulmonary embolism. UTMB uses a specialized algorithm trained to find specific symptoms and flag care teams within seconds of imaging to accelerate treatment.

Like ICAC scoring tools, CNNs are trained in stroke and pulmonary embolism, each receiving a CT scan automatically, looking for indicators such as blocked blood flow and sudden vascular cut-offs.

“Human radiologists can detect these visual properties, but here the detection is automated and occurs in just a few seconds,” McCaffrey said.

CT ordered “under suspicion” of a stroke or pulmonary embolism is automatically sent to the AI. For example, ER clinicians may sag the face or issue “CT stroke” orders, causing algorithms.

Both algorithms include messaging applications that notify the entire care team as soon as discoveries are made. This includes screenshots of images with lateral hair above the lesion location.

“These are specific emergency use cases where the speed at which treatment is important is how quickly and important,” McCaffrey said. “We've seen cases where we can get a few minutes of intervention because we've lifted our heads faster from the AI.”

Reduced hallucinations, anchor bias

To ensure that the model performs as best as possible, UTMB profiles them for sensitivity, specificity, F-1 score, bias, and factors both pre- and repeated post-development.

Thus, for example, the ICAC algorithm is validated prior to deployment by running the model with a balanced CT scan while the radiologist manually scores. Then compare the two. Meanwhile, post-development reviews give radiologists a random subset of CT scans of AI scores, and complete ICAC measurements blinded to AI scores. McCaffrey explained that this allows the team to recur and recur with model errors and detect potential biases (are considered a change in error magnitude and/or direction).

Because they rely so heavily on the initial information encountered by AI and humans, UTMB employs a “peer learning” technique because they lack important details when making decisions to help prevent anchor bias. A random subset of radiation tests is selected, shuffled, anonymized, distributed to various radiologists, and their responses are compared.

Not only does this help to assess the performance of individual radiologists, but it also helps to detect whether the rate of missed findings is higher in studies that have been used to specifically highlight certain abnormalities (and thus leading to bias fixation).

For example, if AI was used to identify and flag x-ray fractures, the team would look at whether fracture flag studies increased the rate of errors in other factors, such as joint spatial stenosis (common in arthritis).

McCaffrey and his team found that consecutive model versions within the class (various versions of GPT-4O) and within the class (GPT-4.5 vs. 3.5) tend to have lower hallucination rates. “But this is non-zero and non-deterministic, so it's fine, but you can't ignore the possibility and impact of hallucinations,” he said.

So they are usually drawn to generation AI tools that do a good job of quoting sources. For example, a model that summarises a patient's medical course while emerging clinical notes that serve as the basis for their output.

“This allows providers to function efficiently as a safeguard against hallucinations,” McCaffrey said.

Flag “basics” to enhance healthcare

UTMB also utilizes AI in several other areas, including automated systems that help determine whether inpatient admissions are justified. The system acts as a co-pilot, automatically extracts all patient notes from the EHR and uses Claude, GPT, and Gemini to summarise and examine them before presenting assessments to staff.

“This allows us to handle the entire patient population and filter/triage patients,” explained McCaffrey. This tool will assist personnel in drafting documents to support admissions and observations.

In other areas, AI is used to review reports such as echocardiology interpretations and clinical notes to identify gaps in care. In many cases, “it's simply flagging the basics,” McCaffrey said.

Healthcare is complex and includes data feeds from anywhere – images, doctor notes, and lab results – not calculated simply because human talent was not sufficient.

This led to what he described as “a massive, large intellectual bottleneck.” Many data are simply not calculated, despite being aggressive and very likely to find things before.

“That's not accusations in a particular location,” McCaffrey emphasized. “It's generally a healthcare state.” Without AI, “we can't deploy intelligence, scrutiny and thought work on the scale necessary to catch everything.”

Daily insights into business use cases in VB every day

If you want to impress your boss, VB Daily has it covered. From regulatory shifts to actual deployments, it provides an internal scoop on what companies are doing with generated AI, allowing you to share the biggest ROI insights.

Please read our privacy policy

Thank you for subscribing. Check out this VB newsletter.

An error has occurred.



Source link

Share. Facebook Twitter Pinterest LinkedIn Tumblr Email
admin
  • Website

Related Posts

VA benefits families and caregivers

June 23, 2025

Expanding Home Health Services and Benefits of Home Health Care | AMA Update Video

June 23, 2025

Shore HomeCare Services continues to provide home care for dementia

June 23, 2025
Leave A Reply Cancel Reply

Top Posts

In an already unstable industry, US musicians struggle for healthcare

July 6, 2025

Understand Latin health at Adelantando Healthcare Conference

January 1, 1970

Seton Hall's new university university program explores women's healthcare for future health leaders

January 1, 1970

North Texas medical chain faces possible $300 million fine for 20,000 fake Medicare claims

January 28, 2003
Don't Miss

Interview: Unison’s departing nurse discusses pay, politics and progress

By adminJuly 5, 2025

The NHS has “taken its eye off the ball” when it comes to safe staffing,…

Community nurses ‘must be central’ in new neighbourhood health service

July 4, 2025

Nurses call for delivery plan to achieve NHS prevention agenda

July 4, 2025

what it means for digital transformation| Nursing Times

July 4, 2025

Subscribe to Updates

Subscribe to our newsletter and never miss our latest news

Subscribe my Newsletter for New Posts & tips Let's stay updated!

About Us
About Us

Welcome to HomeCareNews.us, your trusted source for comprehensive information on home healthcare services. Our mission is to empower individuals and families by providing accurate, up-to-date, and insightful information about essential home care services in USA.

Facebook X (Twitter) Pinterest YouTube WhatsApp
Our Picks

In an already unstable industry, US musicians struggle for healthcare

July 6, 2025

Sign up for healthcare heroes

July 5, 2025

Interview: Unison’s departing nurse discusses pay, politics and progress

July 5, 2025
Most Popular

In an already unstable industry, US musicians struggle for healthcare

July 6, 2025

Understand Latin health at Adelantando Healthcare Conference

January 1, 1970

Seton Hall's new university university program explores women's healthcare for future health leaders

January 1, 1970
  • Home
  • About Us
  • Advertise with Us
  • Contact us
  • DMCA Policy
  • Privacy Policy
  • Terms & Conditions
© 2025 HomecareNews.US

Type above and press Enter to search. Press Esc to cancel.