A gallstones classification method and verification based on deep learning
10.3760/cma.j.cn121382-20240125-00402
- VernacularTitle:基于深度学习的胆结石分类方法研究与验证
- Author:
Qianyun GU
1
;
Chengli SONG
;
Jiawen GUO
;
Dongming YIN
;
Shiju YAN
;
Bo WANG
;
Zhaoyan JIANG
;
Hai HU
Author Information
1. 上海理工大学健康科学与工程学院，教育部微创医疗器械工程研究中心，上海 200093
- Keywords:
Gallstones;
Classification method;
Deep learning;
Accuracy rate;
Precision rate;
Loss value
- From:
International Journal of Biomedical Engineering
2024;47(4):312-317
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To establish and validate a gallstones classification method based on deep learning.Methods:A total of 618 gallstones samples were collected from East Hospital Affiliated to Tongji University, and 1 023 high-definition cross-sectional gallstones profile images were captured to construct a cross-sectional gallstones profile image dataset. Based on the traditional eight-category gallstones classification method, a lightweight network model, MobileNet V3, was trained using deep learning and transfer learning methods. The classification performance of MobileNet was evaluated using a confusion matrix with metrics such as accuracy rate, precision rate, F1 score, and recall rate. The MobileNet V3 was improved and further validated using accuracy and loss values.Results:The accuracy rate (94.17%), precision rate (94.03%), F1 score (92.96%) and recall rate (92.99%) of the improved MobileNet V3 model were better than other networks. The improved MobileNet V3 model achieved the highest accuracy rate (94.17%) in gallstones profile classification and was validated by the test set. The confusion matrix showed a weighted average of accuracy rate (92.0%), precision rate (92.6%), and F1 score (92.2%) for each category of gallstones.Conclusions:Based on deep learning, a high-accuracy gallstones classification method is proposed, which provides a new idea for the intelligent identification of gallstones.
