Epithelial dysplasia is a disorder that is characterized by the differentiation and maturation of a lesion that is produced by the proliferation of abnormal or atypical epithelial cells. These epithelia frequently exhibit architectural and cytological changes, with the majority occurring in squamous cells such as keratinocytes. The accumulation of genetic alterations represents a pivotal factor in the pathogenesis of epithelial dysplasia [1]. The term encompasses lesions in which part or all of the thickness of the epithelium is replaced by cells exhibiting varying degrees of cellular atypia. It is associated with an elevated risk of cancer progression in a range of organs [2–7]. Nevertheless, epithelial dysplasia is a histological phenomenon with no clinical morphological equivalent [1]. In the context of body sites such as the cervix, the term “intraepithelial neoplasia” is used with a similar meaning [8]. This is a histological diagnosis of considerable clinical significance. In studies of oral potentially malignant disorders, Barrett’s esophagus and other pre-cancerous lesions, different grades of epithelial dysplasia demonstrate varying potential for cancer development [3, 5, 6, 9, 10]. There are extensive scientific reports on the clinical management and bench research of epithelial dysplasia. However, the evidence base is weak due to the variation in published studies, particularly in the variability of different grading systems, as well as the inter- and intra-examiner variability, which leads to reliability issues.

Histopathology represents the “gold standard” for the diagnosis of epithelial dysplasia. The interpretation of traditional histomorphology is dependent on the experience of the pathologist, and therefore, more objective and efficient methods are required to integrate and refine diverse morphological, molecular, and multi-omics information. The integration of pathology big data (comprising hundreds of thousands or more pathology images and records) with artificial intelligence (AI) has the potential to facilitate more accurate diagnoses and treatments, reduce the repetitive workload of pathologists, and enable the sharing of pathology diagnostic resources across regions. The advent of digital pathology and the concomitant advancement of computer algorithms have facilitated the application of AI technology in the field of pathology. At present, digital pathology-based AI has demonstrated considerable potential for use in the diagnosis of diseases, the guidance of treatment strategies, the evaluation of therapeutic efficacy, and the prediction of prognosis. This is achieved by the identification of image features, which are automatically learned from digital images and analyzed in relation to labels, such as diagnostic or prognostic purposes [3, 6, 9, 11–13]. These applications of AI in pathology assist in improving the objectivity and accuracy of pathological diagnosis, effectively reducing the workload of pathologists, and laying a solid foundation for the accurate diagnosis and treatment of diseases and the individualized management of patients.

Recent studies conducted comprehensive reviews of the use of AI in managing oral potentially malignant disorders and oral epithelial dysplasia [6, 14]. These findings indicated that AI could serve as a reliable, precise, and personalized medical decision-assisted tool to overcome the limitations of the epithelial dysplasia grading system and contribute to identifying stratified malignant transformation risk and guiding patient treatment strategies [6, 14]. The researches also described the dawn of the application of AI in the field of oral mucosa and explored future implementation pathways and possible challenges. In fact, AI is not only useful for grading management and prognostic assessment of epithelial dysplasia in the oral mucosa but also has promising applications for the clinical management of epithelial dysplasia in multiple organs.

Table 1 presents a review of published reports on the grading and prognostic management of epithelial dysplasia by AI models in various organs, including the oral cavity, larynx, esophagus, stomach, colorectum, and cervix [2–5, 8, 9, 11, 12, 15, 16]. The studies predominantly employed Aperio or Nanozoomer scanning systems with 20× or 40× magnification to obtain digital whole slide images (WSIs). A range of deep learning algorithms, including visual geometry group (VGG) and residual neural network (ResNet), were employed in the construction of AI models. The performance of the models was evaluated using a number of metrics, including the area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, and specificity. The dimensions of a WSI exceed the capacity for training and constructing a deep learning model. To address this issue, multi-instance learning (MIL) methods are employed, whereby it is assumed that at least one small region of the image contains the requisite morphological information for image analysis [17]. Consequently, the entire slide is divided into identical patches of the same size and resolution (224 × 224 pixels or 512 × 512 pixels, etc.) for the creation of deep learning AI. While the size of the patch is reduced, the number of images increases exponentially, resulting in the creation of vast datasets comprising millions or even more patches. The AUC was moderate at the patch level, but integrating pathomics features generated by patch-level models through machine learning or deep learning algorithms could improve WSI-level predictive performance [3, 4]. The results of AI for histological grading assessment and prognostic identification of epithelial dysplasia in all reported organs are promising. This suggests that digital pathology-based AI can facilitate the clinical management of epithelial dysplasia in multiple organs.

It would seem that there is a growing tendency towards the use of computer-assisted digital pathology in place of conventional pathology techniques. The advent of multi-omics and bioinformatics has enabled pathologists to utilize a greater array of molecular and imaging data as diagnostic references. The advent of high-throughput big data has introduced novel professional demands for pathologists. It is becoming increasingly apparent that histopathology slides contain a wealth of information that extends beyond their diagnostic utility. These include molecular alterations, cellular interactions, and prognostic identification. Nevertheless, there is a growing concern regarding the shortage of histopathologists and the increasing workload. In order to address these challenges, it is essential to develop robust, sensitive, and efficient computational tools that can assist in medical decision-making. Such tools may assist pathologists in performing routine diagnostic tasks and alleviating the considerable workloads they currently face. Moreover, they may provide new insights for precision medicine in further research. It seems reasonable to posit that research and translation based on digital pathology AI will prove to be a key strategy for addressing these challenges in the future.