Can Artificial Intelligence Predict The Future?

Published by Daniel at January 1, 2023

Using AI models to predict diabetes

Scientists at Klick Applied Sciences have found a way to change a continuous glucose monitor (CGM) into a robust diabetes screening and prevention tool using artificial intelligence. The Klick organization, established in New York in 1997, works to support commercial actions using its proven business, medical, and specialized expertise. Klick scientists use machine learning and just 12 hours of data from CGMs to decide whether a patient is prediabetic or diabetic. For the study, about 600 patients who were recognized as healthy, prediabetic, or living with type 2 diabetes wore a CGM device for a standard of 12 days. The scientists looked at their glucose measurements over time and developed machine-learning models to see if those values could be used to decide whether that person was healthy, prediabetic, or diabetic.

They discovered their 12-hour model showed similar high accuracy to outcomes from the longer time-frames, correctly identifying two-thirds of patients with prediabetes while demonstrating high precision in identifying healthy patients and those with Type 2 diabetes. The shorter period frame is a big step ahead since most research draws from 10 to 14 days’ worth of readings and often demands analysis from expert clinicians.(1)

For the study, about 600 patients who were recognized as healthy, prediabetic, or living with type 2 diabetes wore a CGM device for a standard of 12 days. The scientists looked at their glucose measurements over time and developed machine-learning models to see if those values could be used to decide whether that person was healthy, prediabetic, or diabetic. They discovered their 12-hour model showed similar high accuracy to outcomes from the longer time-frames, correctly identifying two-thirds of patients with prediabetes while demonstrating high precision in identifying healthy patients and those with Type 2 diabetes. The shorter period frame is a big step ahead since most research draws from 10 to 14 days’ worth of readings and often demands analysis from expert clinicians.(1)

Researchers develop an AI model to predict a person's 10-year risk

Researchers from Massachusetts General Hospital, Boston, USA, developed a deep learning model sample using artificial intelligence (AI) and a simple chest X-ray to assist in predicting a person’s 10-year risk of dying from a stroke or heart attack. The investigation group used a CXR-CVD system “trained” to search around 147,000 chest X-ray image samples from almost 41,000 patients in a cancer screening trial and area patterns associated with cardiovascular conditions. The results were presented in November at the annual Radiological Society of North America (RSNA) meeting. Once developed, the system could be used as a gold standard predictor for a person’s 10-year risk of stroke or heart attack from a single chest X-ray. This study did an excellent job correlating with clinicians’ existing tool, the ASCVD risk score. Although compelling, the new research is preliminary, and more long-term studies are still needed.(2)

The results were presented in November at the annual Radiological Society of North America (RSNA) meeting. Once developed, the system could be used as a gold standard predictor for a person’s 10-year risk of stroke or heart attack from a single chest X-ray. This study did an excellent job correlating with clinicians’ existing tool, the ASCVD risk score. Although compelling, the new research is preliminary, and more long-term studies are still needed.(2)

AI Enables the Largest Brain Tumor Study To-Date

The research laboratory at the Perelman School of Medicine at the University of Pennsylvania, Philadelphia, studies rare conditions like glioblastomas (GBM), an aggressive type of brain tumor. Researchers at Penn University and Intel Corp led the largest-to-date global machine learning step to securely aggregate knowledge from brain scans of 6,314 patients at 71 sites worldwide. This collaboration produced a model that can improve the identification and prediction of boundaries in three tumor sub-compartments; the model followed a staged process. The first stage, an initial public model, was pre-trained using the unique data from the International Brain Tumor Segmentation (BraTS) challenge. The model was tasked with identifying the boundaries of three GBM tumor sub-compartments: Essential compartments are the improved portion, the nucleus of the tumor, and the whole tumor.

This first data of 230 patient cases from 16 different sites and the resulting model was validated. The second stage, called the preliminary consensus example, used the initial public model and included more data from 2,471 patient cases from another 35 sites, which improved its accuracy. The final stage, or final consensus model, used the revised model and included the most significant amount of data from 6,314 patient cases (3,914,680 images) at 71 sites to further optimize and test for generalizability to unseen data. The sum of these three models can help to predict the behavior of such an aggressive and unusual tumor and also helps the clinical decision of doctors to know how to deal with a tumor of these characteristics.(3)

The first stage, an initial public model, was pre-trained using the unique data from the International Brain Tumor Segmentation (BraTS) challenge. The model was tasked with identifying the boundaries of three GBM tumor sub-compartments: Essential compartments are the improved portion, the nucleus of the tumor, and the whole tumor. This first data of 230 patient cases from 16 different sites and the resulting model was validated. The second stage, called the preliminary consensus example, used the initial public model and included more data from 2,471 patient cases from another 35 sites, which improved its accuracy. The final stage, or final consensus model, used the revised model and included the most significant amount of data from 6,314 patient cases (3,914,680 images) at 71 sites to further optimize and test for generalizability to unseen data. The sum of these three models can help to predict the behavior of such an aggressive and unusual tumor and also helps the clinical decision of doctors to know how to deal with a tumor of these characteristics.(3)

AI platform predicts bed state two weeks in advance

The National University Health System, a group of healthcare institutions in Singapore, has recently released a new AI platform that can help to predict hospital bed availability up to a few weeks ahead of time. The software ENDEAVOUR AI platform integrates live data from the next-generation EMR (NGEMR) method to compute multiple AI insights. It hosts various AI tools, one of which can calculate the length of stay of each patient admitted to public hospitals. The AI tool reads patients’ records and doctor notes in real-time, running up to 30 times per hour. It can also deliver clinical signs into factors contributing to a patient’s prolonged stay. The predictive tool is a viable solution to rising bed occupancy rates and increasing bed wait times at emergency rooms. It allows doctors to intervene early in anticipation of problems.

For example, it can flag patients who have been staying past two weeks in the hospital, allowing medical groups to either modify their management or plan a patient’s early transfer to a community hospital for rehabilitation. With its ability to read notes, vital signs, lab test reports, and other parameters, the AI tool can also predict an admitted patient’s risk of deterioration. ENDEAVOUR AI can also automatically alert administrators of climbing ED wait times, allowing workforce resources to be implemented early.(7)

The AI tool reads patients’ records and doctor notes in real-time, running up to 30 times per hour. It can also deliver clinical signs into factors contributing to a patient’s prolonged stay. The predictive tool is a viable solution to rising bed occupancy rates and increasing bed wait times at emergency rooms. It allows doctors to intervene early in anticipation of problems. For example, it can flag patients who have been staying past two weeks in the hospital, allowing medical groups to either modify their management or plan a patient’s early transfer to a community hospital for rehabilitation. With its ability to read notes, vital signs, lab test reports, and other parameters, the AI tool can also predict an admitted patient’s risk of deterioration. ENDEAVOUR AI can also automatically alert administrators of climbing ED wait times, allowing workforce resources to be implemented early.(7)

AI system to predict acute kidney disease 48 hours in advance

Heart failure (HF) is a life-threatening complication of cardiovascular disease. HF patients are more likely to progress to acute kidney injury (AKI) with a poor prognosis. However, it is difficult for doctors to accurately distinguish which patients will develop AKI. This study aimed to construct a machine learning (ML) model to predict AKI occurrence in HF patients. In collaboration with the US Department of Veterans Affairs, researchers used deidentified electronic health record data from more than 700,000 patients collected from 1,200 inpatient and outpatient healthcare facilities to develop the model. The AI system analyzes the patient’s health records, including blood tests, vital signs, and past medical history, and can accurately detect acute kidney injury up to two days earlier than it is currently diagnosed; the data of HF patients from the Medical Information Mart for Intensive Care-IV database were retrospectively analyzed.

An ML model was established to predict AKI development using a decision tree and logistic regression algorithms. Thirty-nine demographic, clinical, and treatment features were used for model establishment. Accuracy, sensitivity, and specificity were used to estimate the performance of the ML algorithms.(4)

The AI system analyzes the patient’s health records, including blood tests, vital signs, and past medical history, and can accurately detect acute kidney injury up to two days earlier than it is currently diagnosed; the data of HF patients from the Medical Information Mart for Intensive Care-IV database were retrospectively analyzed. An ML model was established to predict AKI development using a decision tree and logistic regression algorithms. Thirty-nine demographic, clinical, and treatment features were used for model establishment. Accuracy, sensitivity, and specificity were used to estimate the performance of the ML algorithms.(4)

The AI-powered system can predict the healing of venous leg ulcers

The Royal Melbourne Institute of Technology, a public research University in Melbourne, Australia, has released its latest innovation by researchers, which enabled the identification of chronic leg sores by the second week after the baseline assessment. The team says their latest published results represent a significant leap forward because they permit the identification of these wounds a week before current practices. Their latest clinical study presents AI-powered software to predict how leg ulcers will heal based on thermal images and other features based on the first assessment.

The new method provides details on a wound’s spatial heat distribution and pressure measurement. With 80% accuracy, it predicts whether leg ulcers will heal in 12 weeks without specialized care. Wounds change significantly over the healing trajectory; higher temperatures signal pressure points, potential inflammation, or infection. In comparison, lower temperatures indicate a slower healing rate due to reduced oxygen in the region. The team will also assess whether their method can predict the healing of diabetes-related foot ulcers. Untreated chronic wounds in people with diabetes are the leading cause of limb amputation in Western countries.(5)

Wounds change significantly over the healing trajectory; higher temperatures signal pressure points, potential inflammation, or infection. In comparison, lower temperatures indicate a slower healing rate due to reduced oxygen in the region. The team will also assess whether their method can predict the healing of diabetes-related foot ulcers. Untreated chronic wounds in people with diabetes are the leading cause of limb amputation in Western countries.(5)

AI Clinical Decision Support Tool for Nutrition

A group of researchers from the University of Texas, USA, at Austin’s Dell Medical School, has developed a superior artificial intelligence (AI) clinical decision support tool designed to help clinicians and nutrition employees discuss nutrition plans and engage in shared decision-making with patients regarding diet modifications. Another unique feature of this great software is the anticipation of the patient’s actions and prediction of future deficiencies in the diet. The tool, Nutri, will be first deployed at Lone Star Circle of Care in Austin, Texas. This federally qualified health center focuses on providing care to high-risk populations.

Nutri integrates with the Healthcare EHR system and presents patient diet information to clinicians to support personalized goal setting and progress tracking between visits. It can also deliver clinical signs into factors contributing to a deficient diet. The instrument even drafts potential chart notes to help reduce clinician burden. The device is one of the latest to result from the growing investigation interest in personalized and precision nutrition.(6)

The tool, Nutri, will be first deployed at Lone Star Circle of Care in Austin, Texas. This federally qualified health center focuses on providing care to high-risk populations. Nutri integrates with the Healthcare EHR system and presents patient diet information to clinicians to support personalized goal setting and progress tracking between visits. It can also deliver clinical signs into factors contributing to a deficient diet. The instrument even drafts potential chart notes to help reduce clinician burden. The device is one of the latest to result from the growing investigation interest in personalized and precision nutrition.(6)

References