[The computer-aided diagnosis model of middle ear cholesteatoma based on integrated convolutional neural networks].
Journal:
Zhonghua er bi yan hou tou jing wai ke za zhi = Chinese journal of otorhinolaryngology head and neck surgery
Published Date:
May 7, 2025
Abstract
Middle ear cholesteatoma is a common otolaryngological disease, and traditional diagnostic methods have certain limitations. This study aims to construct a computer-aided diagnosis model for middle ear cholesteatoma based on integrated convolutional neural networks (CNNs) to improve diagnostic accuracy and efficiency. Firstly, Data were collected from patients who visited the Department of Otorhinolaryngology Head and Neck Surgery at the First People's Hospital of Yinchuan between January 2020 and December 2021. 8 000 temporal bone CT images were collected, including 5 000 images diagnosed pathologically as middle ear cholesteatoma and 3 000 normal images. A five-fold cross-validation method was used to divide the dataset into training and testing sets. Next, a transfer learning approach was used to initialize model parameters, and the AlexNet, GoogleNet, and ResNet networks were pre-trained to extract deep features from the images. Then, the Softmax classification algorithm was applied to classify the features, resulting in three independent classifiers. These classifiers were combined using an ensemble learning method with a weighted voting approach to obtain the final diagnostic results. Finally, the model was evaluated by comparing the ensemble classifier with individual classifiers to assess its accuracy, precision, sensitivity, specificity, and diagnostic time, and a comparison with low-mid-and high-experience physician groups was conducted to comprehensively evaluate the model's diagnostic performance. The experimental results showed that the model achieved an accuracy of 88.8%(178/200), precision of 92.9%,(112/120) sensitivity of 89.8%(108/120), and specificity of 88.1%(70/80). The average diagnostic time for individual patient temporal bone CT images was reduced to 2-3 seconds. Compared to the diagnostic results from low-mid-and high-experience physician groups, the model demonstrated significant advantages and effectively assisted clinicians in making rapid and accurate middle ear cholesteatoma diagnoses. The proposed middle ear cholesteatoma diagnostic model based on integrated convolutional neural networks exhibits high recognition accuracy and rapid diagnostic speed, significantly improving clinical diagnostic efficiency, especially in early screening and auxiliary diagnosis, making it of considerable value in clinical practice.
