Categories
Change Password!
Reset Password!
Diagnosing and managing tonsillitis is generally straightforward, but the coronavirus pandemic can heighten the risk of infection for healthcare workers.
Incorporating AI-driven U-Net (U-shaped encoder-decoder network architecture) for automated segmentation enhances the accuracy and efficiency of tonsillitis diagnosis in clinical practice.
Diagnosing and managing tonsillitis is generally straightforward, but the coronavirus pandemic can heighten the risk of infection for healthcare workers. With recent advancements in artificial intelligence (AI), its role in medical imaging has expanded significantly.
This study aimed to explore the best convolutional neural network (CNN) algorithm for correct diagnosis and early therapy of tonsil inflammation or tonsillitis.
To train the CNN, semi-supervised learning with pseudo-labels for self-training was developed, incorporating deep learning algorithms such as U-Net, Pyramid Scene Parsing Network (PSPNet), and Feature Pyramid Network (FPN). The dataset consisted of 485 pharyngoscopic images from 485 participants, including 133 healthy individuals, 295 cases with the common cold, and 57 cases with tonsillitis. For further assessment, the color and texture features from all 485 images were extracted.
As compared to PSPNet and FPN, U-Net was more useful in automatically segmenting oropharyngeal structures with an average Dice coefficient of 97.74% and pixel accuracy of 98.12%, making it highly effective for improving the diagnosis of tonsil inflammation. Normal tonsils typically appeared smooth and uniform with a pinkish hue, closely resembling the surrounding mucosa.
In contrast, tonsillitis, particularly in cases requiring antibiotics, was characterized by white or yellow pus-filled spots or patches and a more granular or uneven texture, signalling inflammation and tissue changes. The U-Net-based algorithm exhibited strong performance with accuracy rates of 93.75%, 97.1% and 91.67% in differentiating the 3 tonsil groups after drilling 485 cases.
U-Net automates the segmentation of oropharyngeal structures, improving feature extraction and enabling accurate AI diagnosis of tonsillitis.
European Achieves of Oto-rhino-laryngology
Harnessing AI for precision tonsillitis diagnosis: a revolutionary approach in endoscopic analysis
Po-Hsuan Jeng et al.
Comments (0)