Abstract:This study developed a new convolutional neural network model to detect and classify gastric lesions as malignant, premalignant, and benign. We used 10,181 white-light endoscopy images from 2606 patients in an 8:1:1 ratio. Lesions were categorized as early gastric cancer (EGC), advanced gastric cancer (AGC), gastric dysplasia, benign gastric ulcer (BGU), benign polyp, and benign erosion. We assessed the lesion detection and classification model using six-class, cancer versus non-cancer, and neoplasm versus non-neoplasm categories, as well as T-stage estimation in cancer lesions (T1, T2-T4). The lesion detection rate was 95.22% (219/230 patients) on a per-patient basis: 100% for EGC, 97.22% for AGC, 96.49% for dysplasia, 75.00% for BGU, 97.22% for benign polyps, and 80.49% for benign erosion. The six-class category exhibited an accuracy of 73.43%, sensitivity of 80.90%, specificity of 83.32%, positive predictive value (PPV) of 73.68%, and negative predictive value (NPV) of 88.53%. The sensitivity and NPV were 78.62% and 88.57% for the cancer versus non-cancer category, and 83.26% and 89.80% for the neoplasm versus non-neoplasm category, respectively. The T stage estimation model achieved an accuracy of 85.17%, sensitivity of 88.68%, specificity of 79.81%, PPV of 87.04%, and NPV of 82.18%. The novel CNN-based model remarkably detected and classified malignant, premalignant, and benign gastric lesions and accurately estimated gastric cancer T-stages.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve accuracy and efficiency in gastric lesion detection. Specifically, a new convolutional neural network (CNN) model has been developed for detecting and classifying gastric lesions, including malignant, potentially malignant, and benign lesions. The study aims to improve the detection rate and classification accuracy of gastric lesions through deep - learning methods, while evaluating the model's ability to predict the depth of invasion of cancerous lesions (T - staging). ### Main Objectives: 1. **Detect and Classify Gastric Lesions**: A multi - category lesion detection model has been developed in the study, which can identify six types of lesions such as early gastric cancer (EGC), advanced gastric cancer (AGC), gastric intraepithelial neoplasia (gastric dysplasia), benign gastric ulcer (BGU), benign polyps, and benign erosions. 2. **Improve Detection Rate and Classification Accuracy**: Evaluate the performance of the model in detecting and classifying different types of gastric lesions, especially the detection of malignant lesions. 3. **Predict the Depth of Invasion of Cancerous Lesions**: Develop a model to predict the T - staging of gastric cancer lesions, thereby helping clinicians decide on treatment options. ### Research Background: - **Early Detection and Diagnosis of Gastric Cancer**: The discovery and accurate diagnosis of early gastric cancer are crucial for the selection of treatment strategies and the improvement of prognosis. High - incidence countries have increased the detection rate of early gastric cancer and reduced the mortality rate of gastric cancer by implementing national screening programs. - **Challenges in Endoscopy**: Accurate identification and diagnosis of gastric lesions require a thorough examination of the stomach, distinguishing abnormal lesions from normal gastric mucosa, and deciding whether to perform a biopsy. These processes are influenced by the experience and expertise of endoscopists and are prone to misdiagnosis. - **Application of Deep Learning**: Deep - learning algorithms are increasingly widely used in the medical field, which can assist clinicians in decision - making, improve the lesion detection rate, and relieve endoscopist fatigue. ### Methods: - **Data Set**: The study used 10,181 white - light endoscopic images of 2,606 patients, which were divided into training, validation, and test sets in the ratio of 8:1:1. - **Model Development**: The YOLOv7 and EfficientNet - B3 models were used for multi - category lesion detection and T - staging classification respectively. - **Performance Evaluation**: Evaluate the performance of the model in terms of detection rate, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). ### Results: - **Lesion Detection Rate**: In the test set, the lesion detection rate of the model was 95.22%, and the detection rates of EGC, AGC, and gastric intraepithelial neoplasia were 100%, 97.22%, and 96.49% respectively. - **Classification Performance**: The accuracy of the six - category classification model was 73.43%, the sensitivity was 80.90%, the specificity was 83.32%, the PPV was 73.68%, and the NPV was 88.53%. - **T - staging Classification**: The accuracy of the T - staging classification model was 85.17%, the sensitivity was 88.68%, the specificity was 79.81%, the PPV was 87.04%, and the NPV was 82.18%. ### Discussion: - **Advantages of the Model**: The model developed in the study performs well in detecting and classifying gastric lesions, especially having a high NPV and sensitivity in distinguishing between malignant and non - malignant lesions. - **Clinical Application**: This model can assist clinicians in making more accurate decisions during endoscopy, reducing the misdiagnosis rate and improving the treatment effect. - **Limitations**: The research data comes from a single institution and requires external validation. In addition, data imbalance may lead to evaluation bias for certain lesion types. In general, the deep - learning model developed in this study performs excellently in gastric lesion detection and classification and has important clinical application value.

Enhanced multi-class pathology lesion detection in gastric neoplasms using deep learning-based approach and validation