Dear editor, Oral squamous cell carcinoma (OSCC) of epithelial origin accounts for 90% of instances of oral cancer, which is the 16th most prevalent malignant neoplasm globally. Oral cancer is preceded by oral potentially malignant diseases (OPMDs). OPMDs are persistent lesions that are clinically evident and have the potential to develop into malignancy1,2. An OPMD is a risk factor for oral cancer. Therefore, a key tactic in controlling oral cancer is the early identification and treatment of OPMDs3,4. There is a window of opportunity for prevention by OPMD therapy since the malignant development of OPMDs may take 5–10 years. Artificial intelligence (AI) has generated a great deal of anticipation over the past decade about its potential applications in the health sciences. In order to assess the diagnostic precision of AI-assisted clinical imaging in the identification of oral cancer and OPMDs, Li et al5 recently conducted a systematic review and meta-analysis. The findings of the meta-analysis, which comprised 17 research in total, showed that AI-based detection utilizing clinical photography has a high diagnostic odds ratio (Fig. 1) and is readily available in the modern day with billions of phone users worldwide. Figure 1.: Forest plots of (A) diagnostic odds ratio, (B) sensitivity, (C) specificity, and (D) negative predictive values for screening oral potentially malignant disorders and oral mucosal cancerous lesions cancerous lesions with influential outliers excluded. However, studies on AI diagnosis of oral mucosal diseases are still very few when compared to AI research in other medical fields, necessitating more thorough research and inquiry. The following are the clinical suggestions and considerations: (1) The current challenge facing research lies in the lack of large-scale, standardized image datasets of oral mucosal lesions. It becomes challenging to compare different AI algorithms, which hinders thorough verification of their precision and applicability. (2) The collection process usually excludes complicated cases, and only few studies take into account the general conditions of the patients. As a result, the reported diagnostic performance during the training and validation phases could not accurately reflect the efficacy in clinical settings. (3) There has not been much study done on uncertainty in the intelligent diagnosis of oral mucosal lesions. Even with high confidence levels, the widely used machine learning algorithm may make incorrect decisions. In other words, individuals are not given clear signs of the ambiguity or ignorance of AI-based diagnosis. Adopting clinically relevant performance criteria and improving the interpretability of AI models are crucial. This strategy will make it easier to apply AI models in therapeutic contexts. Developers should be aware of any potential unintentional algorithmic biases and have a thorough comprehension of the training data. Additionally, external validation is essential to guarantee their applicability to a wide range of people. Significant improvements in diagnosis accuracy are anticipated due to the continuous development of image capturing equipment and the use of various AI algorithms. The use of AI to diagnose a large number of these images is expected to improve healthcare outcomes by enabling precise and effective screening. This study was conducted in compliance with the Transparency in the Reporting of Artificial Intelligence – the TITAN guideline6.
Yang et al. (Tue,) studied this question.