World’s First 6-Lead Consumer ECG; First Device for Predicting Risk of Pressure Ulcers; FDA Cleared FDA Clears Geko Muscle Pump Activator
June 1, 2022Article of the Month – June 2022
June 1, 2022Gastroenterology
The application of AI and ML in gastroenterology is not new; however, advancements are occurring rapidly , and the future landscape of AI driven ...
Artificial intelligence may help lower adenoma miss rates
According to a recent study published in March 2022, using AI during colonoscopies may help decrease adenoma miss rates. This new technology was built in a study consisting of 230 patients who completed different screenings. Of those patients, 116 received screening with new AI, and 114 received a routine screening. The adenoma miss rate was estimated as the number of histologically verified lesions seen during the second colonoscopy divided by the number of lesions detected during both procedures. AI-assisted colonoscopies saw an adenoma miss rate of around 16 percent, compared to 32.4 percent with the routine colonoscopy. Finally, the study concluded that colonoscopies using AI saw a twofold reduction in the miss rate of colorectal neoplasia compared to traditional colonoscopies. (1)
According to a recent study published in March 2022, using AI during colonoscopies may help decrease adenoma miss rates. This new technology was built in a study consisting of 230 patients who completed different screenings. Of those patients, 116 received screening with new AI, and 114 received a routine screening. The adenoma miss rate was estimated as the number of histologically verified lesions seen during the second colonoscopy divided by the number of lesions detected during both procedures.
AI-assisted colonoscopies saw an adenoma miss rate of around 16 percent, compared to 32.4 percent with the routine colonoscopy. Finally, the study concluded that colonoscopies using AI saw a twofold reduction in the miss rate of colorectal neoplasia compared to traditional colonoscopies. (1,2,3)
Artificial Intelligence for Ulcerative Colitis Endoscopy Images
A new study has proposed a novel deep learning convolutional neural network (CNN) architecture to perform more objective and reproducible endoscopic examinations of ulcerative colitis (UC). The researchers from Bærum Hospital in Vestre Viken Health Trust in Norway noted that endoscopic evaluation to grade disease activity reliably, detect complications, including cancer, and verify mucosal healing is needed to care for patients properly. However, such evaluation is hampered by substantial intra-and interobserver variability.(2)
The investigators applied four CNNs to a large subset of 8,000 labeled endoscopic still images from HyperKvasir, the gastrointestinal tract’s largest multi-class image and video data set. All four CNN models achieved extremely high predictive accuracy in all experiments. The researchers plan to explore more extensive and clinically diverse data sets for future studies. They said a more suitable application to CNN would be lower GI tract pathologies during colonoscopies, including diverticulosis, diverticulitis, microscopic colitis, infectious colitis, and pseudomembranous colitis.(2)
A new study has proposed a novel deep learning convolutional neural network (CNN) architecture to perform more objective and reproducible endoscopic examinations of ulcerative colitis (UC). The researchers from Bærum Hospital in Vestre Viken Health Trust in Norway noted that endoscopic evaluation to grade disease activity reliably, detect complications, including cancer, and verify mucosal healing is needed to care for patients properly. However, such evaluation is hampered by substantial intra-and interobserver variability.(2)
The investigators applied four CNNs to a large subset of 8,000 labeled endoscopic still images from HyperKvasir, the gastrointestinal tract’s largest multi-class image and video data set. All four CNN models achieved extremely high predictive accuracy in all experiments. The researchers plan to explore more extensive and clinically diverse data sets for future studies. They said a more suitable application to CNN would be lower GI tract pathologies during colonoscopies, including diverticulosis, diverticulitis, microscopic colitis, infectious colitis, and pseudomembranous colitis.(2)
Advanced Machine Learning to Predict Pancreatic Cancer
In Denmark, scientists created an advanced machine learning (ML) technology to recognize clinical circumstances that led to pancreatic cancer and predict risk over time. This investigation was carried out initially using the Danish National Patient Registry. Researchers noted that there are no reliable biomarkers or screening tools to detect pancreatic cancer early. The current AI method identified a subset of patients with a 25-fold risk for developing pancreatic cancer within three to 36 months through electronic health records. The study used, in part, clinical records comprising 41 years (1977-2018) and 6.1 million patients, of whom roughly 24,000 developed pancreatic cancer. After risk assessment, the investigators tested ML methods to predict cancer occurrence in time intervals of three to 60 months. For cancer occurrence within 36 months, the best ML model showed an odds ratio (OR) of 47.5 for 20% recall and 159.0 for 10% recall.(3)
In Denmark, scientists created an advanced machine learning (ML) technology to recognize clinical circumstances that led to pancreatic cancer and predict risk over time. This investigation was carried out initially using the Danish National Patient Registry. Researchers noted that there are no reliable biomarkers or screening tools to detect pancreatic cancer early. The current AI method identified a subset of patients with a 25-fold risk for developing pancreatic cancer within three to 36 months through electronic health records.
The study used, in part, clinical records comprising 41 years (1977-2018) and 6.1 million patients, of whom roughly 24,000 developed pancreatic cancer. After risk assessment, the investigators tested ML methods to predict cancer occurrence in time intervals of three to 60 months. For cancer occurrence within 36 months, the best ML model showed an odds ratio (OR) of 47.5 for 20% recall and 159.0 for 10% recall.(3)
More straightforward, faster, and more accurate, state of the art of artificial intelligence in GI
Current data suggests that Machine Learning (ML) algorithms in procedures such as upper endoscopy, colonoscopy, and wireless capsule endoscopy (WCE) have transformed clinical practice. They created a new role for augmenting diagnostic methods that could be equal to or better than human physicians. ML models involve image recognition based on computer vision algorithms analyzing data that identify lesions for potential diseases such as Barrett’s Esophagus (BE), H. Pylori Infection, Celiac Disease, and Ulcerative Colitis.
The algorithms can also categorize patients’ conditions, determine their risk, and identify optimal biopsy sites. In addition, these AI models can find early stages of neoplasia with higher specificity and sensitivity rates. For instance, one model identified early gastric cancer and differentiated early gastric cancer with an accuracy of 92.5%, sensitivity of 94%, and specificity of 91%. By combining human expertise with the ML algorithm, identifying the source of gastrointestinal bleeding with the WCE could be possible with higher accuracy (sensitivity of 93% and a specificity of 95%).(4)
Current data suggests that Machine Learning (ML) algorithms in procedures such as upper endoscopy, colonoscopy, and wireless capsule endoscopy (WCE) have transformed clinical practice. They created a new role for augmenting diagnostic methods that could be equal to or better than human physicians. ML models involve image recognition based on computer vision algorithms analyzing data that identify lesions for potential diseases such as Barrett’s Esophagus (BE), H. Pylori Infection, Celiac Disease, and Ulcerative Colitis.
The algorithms can also categorize patients’ conditions, determine their risk, and identify optimal biopsy sites. In addition, these AI models can find early stages of neoplasia with higher specificity and sensitivity rates. For instance, one model identified early gastric cancer and differentiated early gastric cancer with an accuracy of 92.5%, sensitivity of 94%, and specificity of 91%. By combining human expertise with the ML algorithm, identifying the source of gastrointestinal bleeding with the WCE could be possible with higher accuracy (sensitivity of 93% and a specificity of 95%).(4)
AI Generates an Accurate Technique For Detection of Cirrhosis
A fully automated AI-based neural network proved reliable and noninvasive detection of liver cirrhosis. According to the researchers, the algorithms were taught to distinguish patients with cirrhosis from those without it based on electrocardiograms; the specificity and sensitivity in liver transplant patients exceeded 80%. Cirrhosis causes profound changes in the circulatory system that can be detected with abnormalities in ECGs. The AI neural network was trained and then validated on a conventional 12-lead ECG that analyzed heart rhythms from 4,197 patients with cirrhosis within one year of liver transplant and 16,730 age- and sex-matched controls. They have successfully created a fully automated AI model to detect cirrhosis and gauge its severity on 12-lead ECGs.(5)
A fully automated AI-based neural network proved reliable and noninvasive detection of liver cirrhosis. According to the researchers, the algorithms were taught to distinguish patients with cirrhosis from those without it based on electrocardiograms; the specificity and sensitivity in liver transplant patients exceeded 80%. Cirrhosis causes profound changes in the circulatory system that can be detected with abnormalities in ECGs.
The AI neural network was trained and then validated on a conventional 12-lead ECG that analyzed heart rhythms from 4,197 patients with cirrhosis within one year of liver transplant and 16,730 age- and sex-matched controls. They have successfully created a fully automated AI model to detect cirrhosis and gauge its severity on 12-lead ECGs.(5)
Machine Learning in GI Electronic Health Record Data
The Lancet Scientific demonstrated the potential of AI modeling learning-driven outcome prediction in gastroenterology through a model that predicts outcomes following lower GI bleeding. They built an artificial neural network using clinical variables available in the EHR at the initial presentation of lower GI bleed. The program was capable of predicting mortality, recurrent bleeding, and the need for endoscopic intervention with an accuracy higher than 90%.Another ML model utilized 6 parameters (age, baseline hemoglobin, presence of gastric ulcer, gastrointestinal diseases, malignancies, and infections) to predict recurrent peptic ulcer bleeding in 1 year with high accuracy.
The FDA approved a special ANN to predict outcomes in those with primary sclerosing cholangitis. This model employed 9 clinical variables (bilirubin, albumin, alkaline phosphatase, platelets, aspartate aminotransferase, hemoglobin, sodium, patient age, and the number of years since diagnosis) to build a gradient boosted model that predicted hepatic decompensation.(6)
The Lancet Scientific demonstrated the potential of AI modeling learning-driven outcome prediction in gastroenterology through a model that predicts outcomes following lower GI bleeding. They built an artificial neural network using clinical variables available in the EHR at the initial presentation of lower GI bleed. The program was capable of predicting mortality, recurrent bleeding, and the need for endoscopic intervention with an accuracy higher than 90%.
Another ML model utilized 6 parameters (age, baseline hemoglobin, presence of gastric ulcer, gastrointestinal diseases, malignancies, and infections) to predict recurrent peptic ulcer bleeding in 1 year with high accuracy. The FDA approved a special ANN to predict outcomes in those with primary sclerosing cholangitis. This model employed 9 clinical variables (bilirubin, albumin, alkaline phosphatase, platelets, aspartate aminotransferase, hemoglobin, sodium, patient age, and the number of years since diagnosis) to build a gradient boosted model that predicted hepatic decompensation.(6)
CT scans with artificial intelligence, the pancreatic cancer may be predicted years before actual diagnosis
Investigators from the Biomedical Imaging Research Institute at Cedars-Sinai analyzed electronic medical data to determine patients diagnosed with pancreatic cancer within the last 15 years and those who had collected a CT scan years before. In a study measuring the strategy, investigators identified 108 reviewing scans from 72 individuals, including 36 healthy controls. Around 66 scans underwent model development, and 42 venous-phase scans served as external validation. Furthermore, the classifier was tested using 28 scans, including 14 healthy controls and 14 pre-diagnostic scans. The classifier categorized scans into their respective groups with an accuracy of 86%. Researchers noted that the presentation of the system remained consistent during validation with good results, despite the limited amount of data for training.(7)
Investigators from the Biomedical Imaging Research Institute at Cedars-Sinai analyzed electronic medical data to determine patients diagnosed with pancreatic cancer within the last 15 years and those who had collected a CT scan years before. In a study measuring the strategy, investigators identified 108 reviewing scans from 72 individuals, including 36 healthy controls.
Around 66 scans underwent model development, and 42 venous-phase scans served as external validation. Furthermore, the classifier was tested using 28 scans, including 14 healthy controls and 14 pre-diagnostic scans. The classifier categorized scans into their respective groups with an accuracy of 86%. Researchers noted that the presentation of the system remained consistent during validation with good results, despite the limited amount of data for training.(7)
AU Health 1st in state to offer AI-assisted colonoscopy
University Health in Georgia is the first in the state to offer artificial intelligence-assisted colonoscopy. This technology gives doctors a” second opinion” and helps to pinpoint polyps that could be missed during a colonoscopy. The chief of the Gastroenterology Department, Dr. Kenneth Vega, said a typical colonoscopy could miss between five and eight percent of the polyps leading to colon cancer. He mentioned that the new AI technology would cut down on missed polyps and increase the early detection of colon cancer. According to literature, about 15 percent of patients coming in for a screening colonoscopy would have polyps found by this technology and unidentified otherwise.(8)
University Health in Georgia is the first in the state to offer artificial intelligence-assisted colonoscopy. This technology gives doctors a” second opinion” and helps to pinpoint polyps that could be missed during a colonoscopy. The chief of the Gastroenterology Department, Dr. Kenneth Vega, said a typical colonoscopy could miss between five and eight percent of the polyps leading to colon cancer.
He mentioned that the new AI technology would cut down on missed polyps and increase the early detection of colon cancer. According to literature, about 15 percent of patients coming in for a screening colonoscopy would have polyps found by this technology and unidentified otherwise.(8)