Editorial: Early Diagnosis in Head and Neck Cancer: Advances, Techniques, and Challenges

Natalia Koerich Laureano^1*Elena Riet Correa Rivero²Giuseppe Pannone³Fernanda Visioli¹

• ¹Federal University of Rio Grande do Sul, Porto Alegre, Brazil
• ²Federal University of Santa Catarina, Florianópolis, Brazil
• ³Institute of Ophthalmology, University of Foggia, Foggia, Italy

The development of artificial intelligence tools to aid clinical, image, and histopathological diagnosis is becoming a reality in cancer research. However, the previously developed methods still lack a good accuracy for clinical application. Zhang et al. proposed a radiomics and deep learning (DL) fusion model based on MRI to distinguish sinonasal squamous cell carcinoma (SNSCC) and sinonasal lymphoma (SNL), two malignant neoplasms with similar manifestations and imaging characteristics. The combined model showed a higher accuracy to distinguish SNSCC and SNL, presenting itself as a non-invasive tool, capable of predicting the diagnosis and helping to determine personalized treatment plan.Despite increased access to imaging tests such as ultrasound (US) and computed tomography (CT), some rare cancers still face misdiagnosis. Yu et al. presented a rare case of thyroid squamous cell carcinoma clinically presented as a painful neck mass, diagnosed prior as a cystic solid mass after US, CT and laryngoscopic screening, and treated as subacute thyroiditis. Therefore, the diagnosis of thyroid nodules can be challenging.The Thyroid Imaging, Reporting and Data System (TIRADS) is an ultrasound-based risk stratification system to aid in differentiating between benign and malignant thyroid nodules. However, some stratifications still lack precision in diagnosing malignancy, with a high rate of variation (2). The development of supplementary technologies to enhance accuracy of thyroid nodules image diagnosis might help clinical decision-making and reduce misdiagnosis. Xie et al. investigated the efficacy of Contrast-Enhanced Ultrasound (CEUS), which assesses the microvascular distribution of lesions and parenchymal perfusion, qualitative and quantitative parameters through multifactor analysis for differentiation of small solid C-TIRADS 4 thyroid nodules. The logistic model demonstrated significantly higher diagnostic performance compared to individual parameters, with accuracy of 90%. The combination of C-TIRADS with qualitative and quantitative CEUS parameters increases the diagnostic accuracy of malignant thyroid nodules.To improve the diagnosis of thyroid adenomatoid nodules by ultrasound (TANU), Cheng et al. developed a nomogram that incorporates US-based radiomic features and clinical information. A radiomic nomogram, a clinical nomogram, and a radiomic-clinical nomogram were constructed using machine learning algorithms. The nomogram that incorporated clinical data and radiomic features showed higher accuracy in diagnosing TANU and guiding therapeutic decisions.Papillary thyroid carcinoma (PTC), the most common TC subtype, generally has a favorable prognosis, however up to 40% may develop lymph node metastasis (LNM). The LNM screening is performed by the US, although their anatomical characteristics difficult the US diagnosis sensitivity. Thus, Hao et al. proposed a clinical prediction model that integrates ultrasound features, clinical parameters, and biochemical markers to provide a practical tool to identify patients at risk of LNM and guide surgical treatment strategies. Regression models identified age, male sex, nodule size, multifocal lesions, capsular contact or invasion and illdefined margins as risk variables. A nomogram was created based on those risk factors. Each risk factor produced a cumulative total, which corresponds to a specific value on the risk axis, indicating the likelihood of LNM in patients with PTC. A high score indicated an increased risk of LNM for patients with PTC. The prediction model showed excellent accuracy and can guide clinical decision-making.The present research topic highlights novel approaches in HNC early diagnosis and noninvasive techniques with promising results for clinical application in the future. The prospect of future applicability of noninvasive techniques, such as liquid saliva biopsy, is an important step towards early diagnosis and potentially reducing the high local recurrence ratecurrently around 50% within two years -which remains a major obstacle to improving patient outcomes. Furthermore, when combined with next-generation sequencing, enable the identification of new biomarkers to monitor tumor progression and screen patients with a poorer prognosis.The use of imaging tests is essential for the diagnosis of cancers in anatomical locations that preclude direct visual inspection during clinical care. The use of different imaging diagnostics, enhanced by the use of DL and radiomics, has facilitated the identification of parameters previously imperceptible in human evaluation. The proposed imaging prediction models integrate multiple approaches, underscoring the importance of combining diverse evaluation methods and artificial intelligence to enhance diagnostic accuracy and clinical applicability. Their applicability can reduce incorrect diagnoses and unnecessary additional examinations. Moreover, the combination of these findings with clinical information proposes clinical prediction models as a possible tool to determine the diagnosis based on information from imaging exams and to screen patients at risk of developing metastases.