Emerging applications of artificial intelligence in the diagnosis and risk stratification of head and neck cancer: a scoping review

Cancer of the oral cavity, oropharynx, hypopharynx, larynx, and related anatomical structures constitutes a global public health problem. These neoplasms are associated with a high morbidity and mortality rate, functional limitations, and impaired quality of life, particularly in advanced stages, underscoring the need for early diagnosis and risk stratification from the initial phases of the disease (1,2). Despite the implementation of surgical interventions, radiotherapy, chemotherapy, and immunotherapy, the prognosis of head and neck cancer remains highly variable depending on a range of clinical, pathological, and biological factors. Tumor heterogeneity, etiological variations such as the human papillomavirus, and variability in therapeutic response make it impossible to accurately predict clinical outcomes using conventional techniques (3,4). In the current scenario, appropriate risk stratification is essential to optimize clinical decision-making and advance toward more individualized medicine. In recent years, artificial intelligence (AI) has sparked growing interest in the field of oncology due to its ability to analyze significant volumes of complex, multidimensional data. In the field of oncology, machine learning and deep learning models have optimized diagnosis, tumor classification, and prognosis, in some instances surpassing conventional statistical models (5,6). These technologies have the ability to integrate clinical, radiological, histopathological, and molecular data to provide a more comprehensive representation of tumor behavior. The implementation of techniques such as radiomics, which enable the extraction of quantitative features from medical images, has made it possible to characterize tumors non-invasively. Other studies have shown that radiomic and deep learning models can identify patterns associated with recurrence, survival, and treatment response in head and neck cancer, providing prognostic information that is imperceptible to the human eye (7–9). These methodologies have solidified the postulate that medical images contain latent information of significant clinical value. Additionally, artificial intelligence has been employed in the analysis of digital histopathological images and genomic data, leading to more precise tumor classification and the identification of risk groups with implications for treatment. The integration of multimodal data using advanced algorithms represents one of the most promising fields in precision oncology (10–12). However, despite its rapid development, the existing evidence exhibits notable heterogeneity in terms of methodologies, data sources, performance metrics, and validation levels. Numerous studies are based on retrospective cohorts, small sample sizes, or the absence of external validation, factors that raise questions about the reproducibility and clinical applicability of the derived models (13–15). Unaddressed challenges persist, such as the risk of algorithmic bias, model interpretability, and their integration into clinical workflows. In the current scenario, it is imperative to develop a systematic and comprehensive synthesis that elucidates the current state of the science, emerging trends, and key methodological shortcomings in the implementation of artificial intelligence for the diagnosis and risk stratification of head and neck cancer. Given the volume, diversity, and exploratory nature of the available evidence, a scoping review is best suited to address this question (16–18). Consequently, the purpose of this scoping review is to map and characterize the contemporary evidence regarding the emerging use of artificial intelligence for the diagnosis and risk stratification of head and neck cancer. The methodological approaches, data types, quantified results, and identified limitations will be detailed to guide future research and facilitate a responsible, evidence-based clinical implementation (19,20).