Related Experiment Video
Updated: Jun 25, 2025
03:05
Author Spotlight: Advancing Early Detection and Treatment of Gastrointestinal Tumors
Published on: February 16, 2024
1.0K
Enhanced multi-class pathology lesion detection in gastric neoplasms using deep learning-based approach and
Byeong Soo Kim1, Bokyung Kim2, Minwoo Cho3
1Interdisciplinary Program in Bioengineering, Graduate School, Seoul National University, Seoul, 08826, Korea.
Scientific Reports
|May 21, 2024
Summary
A new AI model accurately detects and classifies gastric lesions, distinguishing between malignant, premalignant, and benign conditions. This advanced system also estimates gastric cancer T-stages, improving diagnostic capabilities.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Gastroenterology
Background:
- Gastric lesion diagnosis relies on endoscopy, but accurate classification can be challenging.
- Early and accurate detection of gastric cancer and precancerous lesions is crucial for patient outcomes.
Purpose of the Study:
- To develop and evaluate a novel convolutional neural network (CNN) model for detecting and classifying gastric lesions.
- To assess the model's ability to differentiate between malignant, premalignant, and benign gastric conditions.
- To evaluate the model's performance in estimating the T-stage of gastric cancer.
Main Methods:
- A dataset of 10,181 white-light endoscopy images from 2,606 patients was utilized.
- A CNN model was trained to detect and classify lesions into six categories: early gastric cancer (EGC), advanced gastric cancer (AGC), gastric dysplasia, benign gastric ulcer (BGU), benign polyp, and benign erosion.
- Model performance was assessed using various metrics including accuracy, sensitivity, specificity, PPV, and NPV for different classification tasks and T-stage estimation.
Main Results:
- The model achieved a high lesion detection rate of 95.22% on a per-patient basis.
- For the six-class classification, accuracy was 73.43%, sensitivity 80.90%, and NPV 88.53%.
- The T-stage estimation model demonstrated 85.17% accuracy, 88.68% sensitivity, and 82.18% NPV.
Conclusions:
- The developed CNN model shows significant potential for accurately detecting and classifying various gastric lesions.
- The model's ability to estimate gastric cancer T-stages offers valuable information for clinical decision-making.
- This AI-driven approach can aid endoscopists in improving the diagnosis of gastric pathologies.
