Spotting malignancies from gastric endoscopic images using deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Gastric cancer affects over one million people globally each year.
• Endoscopic examination is crucial for early detection, but repetitive visual analysis can lead to missed diagnoses.
• Automated systems are needed to aid medical professionals in identifying gastrointestinal abnormalities.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying benign ulcers and gastric cancer from endoscopic images.
• To improve the accuracy and efficiency of malignancy detection in gastrointestinal endoscopy.

Main Methods:

• Utilized a dataset of 200 normal, 367 cancer, and 220 ulcer endoscopic images.
• Applied deep neural network models (Inception, ResNet, VGGNet) using a transfer-learning approach.
• Conducted 100 experiments with data partitioning and model building, averaging performance metrics.

Main Results:

• Achieved high area under the curve (AUC) values for classifiers: 0.95 (normal vs. cancer), 0.97 (normal vs. ulcer), and 0.85 (ulcer vs. cancer).
• ResNet model demonstrated superior performance.
• Accuracies exceeded 90% for classifications involving normal tissue; ulcer vs. cancer classification achieved 77.1% accuracy.

Conclusions:

• The proposed deep learning method shows promising results for clinical application in gastric cancer detection.
• Automated classification can serve as a valuable adjunct to manual endoscopic review, reducing the risk of missed diagnoses.
• This technology can enhance diagnostic accuracy and patient care in gastroenterology.