Основні поняття
A deep learning model called Masked Latent Transformer with Random Masking Ratio (MLTrMR) is proposed to advance the automated diagnosis of dental fluorosis, a chronic disease caused by long-term overconsumption of fluoride.
Анотація
The authors construct the first open-source dental fluorosis image dataset (DFID) to lay the foundation for deep learning research in this field. They propose a pioneering deep learning model called MLTrMR, which introduces a mask latent modeling scheme based on Vision Transformer to enhance contextual learning of dental fluorosis lesion characteristics.
MLTrMR consists of a latent embedder, encoder, and decoder. The latent embedder extracts latent tokens from the original image, while the encoder and decoder, comprising the latent transformer (LT) block, process unmasked tokens and predict masked tokens, respectively. To mitigate the lack of inductive bias in Vision Transformer, the LT block incorporates latent tokens to enhance the learning capacity of latent lesion features.
The authors design an auxiliary loss function to constrain the parameter update direction of MLTrMR by reshaping the decoder output into a feature map matching the shape of the original image. This reduces the discrepancy between the feature map and the original image, guiding the model towards optimal parameter updates and significantly improving performance.
The authors create four model variants to investigate the impact of various hyperparameters on MLTrMR. On the DFID dataset, MLTrMR achieves an accuracy of 80.19%, an F1 score of 75.79%, and a quadratic weighted kappa of 81.28%, making it state-of-the-art for automated diagnosis of dental fluorosis.
Статистика
Dental fluorosis is a chronic disease caused by long-term overconsumption of fluoride, leading to changes in the appearance of tooth enamel.
Dental fluorosis is widespread worldwide, with China and India being the most affected countries.
Even dental professionals may not be able to accurately distinguish dental fluorosis and its severity based on tooth images.
Цитати
"Dental fluorosis can have a range of effects, from aesthetic concerns to negative impacts on mental and physical well-being. Therefore, early detection of dental fluorosis is essential for effective prevention and treatment."
"As of April 2024, the literature search on the Web of Science reveals only four studies on the aided diagnosis of dental fluorosis."