User login
AI Tool Helps Patients Assess Bowel Preparation for Colonoscopy
TOPLINE:
An artificial intelligence (AI) model, developed using stool images, accurately assessed whether a patient’s bowel preparation was sufficient for colonoscopy. The best version of the model achieved an area under the receiver operating characteristic curve (AUC) of 0.95, accuracy of 0.90, sensitivity of 0.93, and specificity of 0.86.
METHODOLOGY:
- Patients often need help during bowel preparation for colonoscopy, which increases staff workload; up to 20%-25% of colonoscopies are reported to be inadequately prepared. Researchers developed and tested an AI tool (AI-PREPOO) using stool images to assess whether patients were ready for colonoscopy.
- They conducted a multicenter observational study in Japan between 2022 and 2023 that included 37 patients scheduled for colonoscopy (median age, 57 years; 45.9% women).
- After starting consumption of a 2-liter polyethylene glycol solution, patients used smartphones to take photos of their stool in the toilet after each bowel movement and uploaded the images to a secure web server.
- The images were divided into training and test sets. Images were classified as “ready” for colonoscopy when the stool was clear or light yellow and watery with no solid content.
- Four image-recognition models based on different deep learning architectures were developed using transfer learning to classify readiness for colonoscopy.
TAKEAWAY:
- Researchers collected 282 stool images, with 141 classified as ready and 141 as not ready. Of these, 224 images were used for training (the number augmented to 2240 images) and 58 for testing.
- All four AI-PREPOO models showed high performance, with AUCs ranging from 0.92 to 0.95; pairwise differences in AUCs were not significant.
- The AI-PREPOO 1 model, based on the MobileNetV3-Small architecture, showed the most balanced performance, with an AUC of 0.95, accuracy of 0.90, sensitivity of 0.93, and specificity of 0.86 on the test set.
- During colonoscopy, all patients had a Boston Bowel Preparation Scale score of 6 or higher, indicating that “ready” images corresponded to an adequately prepared bowel.
IN PRACTICE:
“If implemented as a mobile application, our model would allow patients to quickly and independently assess bowel preparation adequacy, reducing reliance on nurses and alleviating embarrassment associated with sharing stool images. This approach could also lessen nurses’ workload by minimizing unnecessary inquiries and preventing excessive or insufficient bowel preparation due to uncertainty,” the authors wrote.
SOURCE:
This study was led by Kosuke Kojima, Graduate School of Medical and Dental Sciences, Niigata University in Niigata, Japan. It was published online in the Journal of Gastroenterology and Hepatology.
LIMITATIONS:
The small dataset limited generalizability and increased the risk for overfitting. In real-world practice, stool images might vary in lighting, angle, focus, zoom, and background; thus, a larger and more diverse dataset was needed. The model lacked external validation in an independent or prospective cohort.
DISCLOSURES:
This study received support from the Japanese Foundation for Research and Promotion of Endoscopy. The authors reported having no conflicts of interest.
This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.
A version of this article first appeared on Medscape.com.
TOPLINE:
An artificial intelligence (AI) model, developed using stool images, accurately assessed whether a patient’s bowel preparation was sufficient for colonoscopy. The best version of the model achieved an area under the receiver operating characteristic curve (AUC) of 0.95, accuracy of 0.90, sensitivity of 0.93, and specificity of 0.86.
METHODOLOGY:
- Patients often need help during bowel preparation for colonoscopy, which increases staff workload; up to 20%-25% of colonoscopies are reported to be inadequately prepared. Researchers developed and tested an AI tool (AI-PREPOO) using stool images to assess whether patients were ready for colonoscopy.
- They conducted a multicenter observational study in Japan between 2022 and 2023 that included 37 patients scheduled for colonoscopy (median age, 57 years; 45.9% women).
- After starting consumption of a 2-liter polyethylene glycol solution, patients used smartphones to take photos of their stool in the toilet after each bowel movement and uploaded the images to a secure web server.
- The images were divided into training and test sets. Images were classified as “ready” for colonoscopy when the stool was clear or light yellow and watery with no solid content.
- Four image-recognition models based on different deep learning architectures were developed using transfer learning to classify readiness for colonoscopy.
TAKEAWAY:
- Researchers collected 282 stool images, with 141 classified as ready and 141 as not ready. Of these, 224 images were used for training (the number augmented to 2240 images) and 58 for testing.
- All four AI-PREPOO models showed high performance, with AUCs ranging from 0.92 to 0.95; pairwise differences in AUCs were not significant.
- The AI-PREPOO 1 model, based on the MobileNetV3-Small architecture, showed the most balanced performance, with an AUC of 0.95, accuracy of 0.90, sensitivity of 0.93, and specificity of 0.86 on the test set.
- During colonoscopy, all patients had a Boston Bowel Preparation Scale score of 6 or higher, indicating that “ready” images corresponded to an adequately prepared bowel.
IN PRACTICE:
“If implemented as a mobile application, our model would allow patients to quickly and independently assess bowel preparation adequacy, reducing reliance on nurses and alleviating embarrassment associated with sharing stool images. This approach could also lessen nurses’ workload by minimizing unnecessary inquiries and preventing excessive or insufficient bowel preparation due to uncertainty,” the authors wrote.
SOURCE:
This study was led by Kosuke Kojima, Graduate School of Medical and Dental Sciences, Niigata University in Niigata, Japan. It was published online in the Journal of Gastroenterology and Hepatology.
LIMITATIONS:
The small dataset limited generalizability and increased the risk for overfitting. In real-world practice, stool images might vary in lighting, angle, focus, zoom, and background; thus, a larger and more diverse dataset was needed. The model lacked external validation in an independent or prospective cohort.
DISCLOSURES:
This study received support from the Japanese Foundation for Research and Promotion of Endoscopy. The authors reported having no conflicts of interest.
This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.
A version of this article first appeared on Medscape.com.
TOPLINE:
An artificial intelligence (AI) model, developed using stool images, accurately assessed whether a patient’s bowel preparation was sufficient for colonoscopy. The best version of the model achieved an area under the receiver operating characteristic curve (AUC) of 0.95, accuracy of 0.90, sensitivity of 0.93, and specificity of 0.86.
METHODOLOGY:
- Patients often need help during bowel preparation for colonoscopy, which increases staff workload; up to 20%-25% of colonoscopies are reported to be inadequately prepared. Researchers developed and tested an AI tool (AI-PREPOO) using stool images to assess whether patients were ready for colonoscopy.
- They conducted a multicenter observational study in Japan between 2022 and 2023 that included 37 patients scheduled for colonoscopy (median age, 57 years; 45.9% women).
- After starting consumption of a 2-liter polyethylene glycol solution, patients used smartphones to take photos of their stool in the toilet after each bowel movement and uploaded the images to a secure web server.
- The images were divided into training and test sets. Images were classified as “ready” for colonoscopy when the stool was clear or light yellow and watery with no solid content.
- Four image-recognition models based on different deep learning architectures were developed using transfer learning to classify readiness for colonoscopy.
TAKEAWAY:
- Researchers collected 282 stool images, with 141 classified as ready and 141 as not ready. Of these, 224 images were used for training (the number augmented to 2240 images) and 58 for testing.
- All four AI-PREPOO models showed high performance, with AUCs ranging from 0.92 to 0.95; pairwise differences in AUCs were not significant.
- The AI-PREPOO 1 model, based on the MobileNetV3-Small architecture, showed the most balanced performance, with an AUC of 0.95, accuracy of 0.90, sensitivity of 0.93, and specificity of 0.86 on the test set.
- During colonoscopy, all patients had a Boston Bowel Preparation Scale score of 6 or higher, indicating that “ready” images corresponded to an adequately prepared bowel.
IN PRACTICE:
“If implemented as a mobile application, our model would allow patients to quickly and independently assess bowel preparation adequacy, reducing reliance on nurses and alleviating embarrassment associated with sharing stool images. This approach could also lessen nurses’ workload by minimizing unnecessary inquiries and preventing excessive or insufficient bowel preparation due to uncertainty,” the authors wrote.
SOURCE:
This study was led by Kosuke Kojima, Graduate School of Medical and Dental Sciences, Niigata University in Niigata, Japan. It was published online in the Journal of Gastroenterology and Hepatology.
LIMITATIONS:
The small dataset limited generalizability and increased the risk for overfitting. In real-world practice, stool images might vary in lighting, angle, focus, zoom, and background; thus, a larger and more diverse dataset was needed. The model lacked external validation in an independent or prospective cohort.
DISCLOSURES:
This study received support from the Japanese Foundation for Research and Promotion of Endoscopy. The authors reported having no conflicts of interest.
This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.
A version of this article first appeared on Medscape.com.
AI Tool Helps Patients Assess Bowel Preparation for Colonoscopy
AI Tool Helps Patients Assess Bowel Preparation for Colonoscopy