AI Device for Classifying Stomach Lesions in Real-Time Endoscopy
Author Information
Author(s): Gong Eun Jeong, Bang Chang Seok, Lee Jae Jun
Primary Institution: Hallym University College of Medicine
Hypothesis
Can an edge computing device using AI accurately classify gastric neoplasms in real-time during endoscopy?
Conclusion
The study successfully developed an AI model that accurately classifies gastric lesions in real-time with high accuracy.
Supporting Evidence
- The AI model achieved an internal-test accuracy of 93.8%.
- The prospective-test accuracy was 93.3%.
- The model's inference speed was 2–3 ms on GPU and 5–6 ms on CPU.
- Expert endoscopists reported no delays or interference from the AI model during exams.
Takeaway
This study created a smart device that helps doctors quickly identify stomach problems during exams, making it easier to catch issues early.
Methodology
The study used 15,910 endoscopic images to train an AI model, which was then tested for accuracy in real-world procedures.
Potential Biases
Potential biases may arise from using data from only one medical institution.
Limitations
The model's generalizability is limited as it was trained on data from a single institution, and it requires high-performance hardware.
Participant Demographics
Patients with gastric neoplasms confirmed through endoscopy at a single hospital.
Statistical Information
Confidence Interval
93.4–94.2%
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website