Artificial Intelligence-Driven Radiomics in Head and Neck Cancer: Current Status and Future Prospects.

BACKGROUND: Radiomics is a rapidly growing field used to leverage medical radiological images by extracting quantitative features. These are supposed to characterize a patient's phenotype, and when combined with artificial intelligence techniques, to improve the accuracy of diagnostic models and clinical outcome prediction. OBJECTIVES: This review aims at examining the application areas of artificial intelligence-based radiomics (AI-based radiomics) for the management of head and neck cancer (HNC). It further explores the workflow of AI-based radiomics for personalized and precision oncology in HNC. Finally, it examines the current challenges of AI-based radiomics in daily clinical oncology and offers possible solutions to these challenges. METHODS: Comprehensive electronic databases (PubMed, Medline via Ovid, Scopus, Web of Science, CINAHL, and Cochrane Library) were searched following the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines. The quality of included studies and their risk of biases were evaluated using the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD)and Prediction Model Risk of Bias Assessment Tool (PROBAST). RESULTS: Out of the 659 search hits retrieved, 45 fulfilled the inclusion criteria. Our review revealed that the application of AI-based radiomics model as an ancillary tool for improved decision-making in HNC management includes radiomics-based cancer diagnosis and radiomics-based cancer prognosis. The radiomics-based cancer diagnosis includes tumor staging, tumor grading, and classification of malignant and benign tumors. Similarly, radiomics-based cancer prognosis includes prediction for treatment response, recurrence, metastasis, and survival. In addition, the challenges in the implementation of these models for clinical evaluations include data imbalance, feature engineering (extraction and selection), model generalizability, multi-modal fusion, and model interpretability. CONCLUSION: Considering the highly subjective and interobserver variability that is peculiar to the interpretation of medical images by expert clinicians, AI-based radiomics seeks to offer potentially useful quantitative information, which is not visible to the human eye or unintentionally often remain ignored during clinical imaging practice. By enabling the extraction of this type of information, AI-based radiomics has the potential to revolutionize HNC oncology, providing a platform for more personalized, higher quality, and cost-effective care for HNC patients.

Genome-wide somatic mutation analysis of sinonasal adenocarcinoma with and without wood dust exposure.

BACKGROUND: Sinonasal adenocarcinoma is a rare cancer, encompassing two different entities, the intestinal-type sinonasal adenocarcinoma (ITAC) and the non-intestinal-type sinonasal adenocarcinoma (non-ITAC). Occurrence of ITAC is strongly associated with exposure to hardwood dusts. In countries with predominant exposure to softwood dust the occurrence of sinonasal adenocarcinomas is lower and the relative amount of non-ITACs to ITACs is higher. The molecular mechanisms behind the tumorigenic effects of wood dust remain largely unknown. METHODS: We carried out whole-genome sequencing of formalin-fixed paraffin-embedded (FFPE) samples of sinonasal adenocarcinomas from ten wood dust-exposed and six non-exposed individuals, with partial tobacco exposure data. Sequences were analyzed for the presence of mutational signatures matching COSMIC database signatures. Driver mutations and CN variant regions were characterized. RESULTS: Mutation burden was higher in samples of wood dust-exposed patients (p = 0.016). Reactive oxygen species (ROS) damage-related mutational signatures were almost exclusively identified in ITAC subtype samples (p = 0.00055). Tobacco smoke mutational signatures were observed in samples of patients with tobacco exposure or missing information, but not in samples from non-exposed patients. A tetraploidy copy number (CN) signature was enriched in ITAC subtype (p = 0.042). CN variation included recurrent gains in COSMIC Cancer Gene Census genes TERT, SDHA, RAC1, ETV1, PCM1, and MYC. Pathogenic variants were observed most frequently in TP53, NF1, CHD2, BRAF, APC, and LRP1B. Driver mutations and copy number gains did not segregate by subtype. CONCLUSIONS: Our analysis identified distinct mutational characteristics in ITAC and non-ITAC. Mutational signature analysis may eventually become useful for documentation of occupation-related cancer, while the exact mechanisms behind wood dust-driven carcinogenesis remain elusive. The presence of homologous recombination deficiency signatures implies a novel opportunity for treatment, but further studies are needed.

Metastatic cutaneous squamous cell carcinoma accounts for nearly all squamous cell carcinomas of the parotid gland.

Primary squamous cell carcinoma of the parotid gland (pSCCP) has long been recognized as a separate entity and is included in the WHO classifications of salivary gland tumors. However, it is widely accepted among head and neck pathologists that pSCCP is exceptionally rare. Yet, there are many publications describing series of pSCCP and data from SEER and other cancer register databases indicate erroneously an increasing incidence of pSCCP. Importantly, pSCCP and metastatic (secondary) squamous cell carcinoma to the parotid gland (mSCCP) have nearly identical histological features, and the diagnosis of pSCCP should only be made after the exclusion of mSCCP. Moreover, all of the histological diagnostic criteria proposed to be in favor of pSCCP (such as, for example, dysplasia of ductal epithelium) can be encountered in unequivocal mSCCP, thereby representing secondary growth along preexistent ducts. Squamous cell differentiation has also been reported in rare genetically defined primary parotid carcinomas, either as unequivocal histological squamous features (e.g., NUT carcinoma, mucoepidermoid carcinoma), by immunohistochemistry (e.g., in NUT carcinoma, adamantinoma-like Ewing sarcoma, basal-type salivary duct carcinoma, mucoepidermoid carcinoma), or a combination of both. Another major issue in this context is that the International Classification of Diseases (ICD) coding system does not distinguish between primary or metastatic disease, resulting in a large number of patients with mSCCP being misclassified as pSCCP. Immunohistochemistry and new molecular biomarkers have significantly improved the accuracy of the diagnosis of many salivary gland neoplasms, but until recently there were no biomarkers that can accurately distinguish between mSCCP and pSCCP. However, recent genomic profiling studies have unequivocally demonstrated that almost all SCCP analyzed to date have an ultraviolet light (UV)-induced mutational signature typical of mSCCP of skin origin. Thus, mutational signature analysis can be a very useful tool in determining the cutaneous origin of these tumors. Additional molecular studies may shed new light on this old diagnostic and clinical problem. This review presents a critical view of head and neck experts on this topic.

Assessment of targeted therapy opportunities in sinonasal cancers using patient-derived functional tumor models.

Malignant tumors derived from the epithelium lining the nasal cavity region are termed sinonasal cancers, a highly heterogeneous group of rare tumors accounting for 3 - 5 % of all head and neck cancers. Progress with next-generation molecular profiling has improved our understanding of the complexity of sinonasal cancers and resulted in the identification of an increasing number of distinct tumor entities. Despite these significant developments, the treatment of sinonasal cancers has hardly evolved since the 1980s, and an advanced sinonasal cancer presents a poor prognosis as targeted therapies are usually not available. To gain insights into potential targeted therapeutic opportunities, we performed a multiomics profiling of patient-derived functional tumor models to identify molecular characteristics associated with pharmacological responses in the different subtypes of sinonasal cancer. METHODS: Patient-derived ex vivo tumor models representing four distinct sinonasal cancer subtypes: sinonasal intestinal-type adenocarcinoma, sinonasal neuroendocrine carcinoma, sinonasal undifferentiated carcinoma and SMARCB1 deficient sinonasal carcinoma were included in the analyses. Results of functional drug screens of 160 anti-cancer therapies were integrated with gene panel sequencing and histological analyses of the tumor tissues and the ex vivo cell cultures to establish associations between drug sensitivity and molecular characteristics including driver mutations. RESULTS: The different sinonasal cancer subtypes display considerable differential drug sensitivity. Underlying the drug sensitivity profiles, each subtype was associated with unique molecular features. The therapeutic vulnerabilities correlating with specific genomic background were extended and validated with in silico analyses of cancer cell lines representing different human cancers and with reported case studies of sinonasal cancers treated with targeted therapies. CONCLUSION: The results demonstrate the importance of understanding the differential biology and the molecular features associated with the different subtypes of sinonasal cancers. Patient-derived ex vivo tumor models can be a powerful tool for investigating these rare cancers and prioritizing targeted therapeutic strategies for future clinical development and personalized medicine.

Molecularly defined sinonasal malignancies: an overview with focus on the current WHO classification and recently described provisional entities.

Classification of tumors of the head and neck has evolved in recent decades including a widespread application of molecular testing in tumors of the sinonasal tract, salivary glands, and soft tissues with a predilection for the head and neck. The availability of new molecular techniques has allowed for the definition of multiple novel tumor types unique to head and neck sites. Moreover, an expanding spectrum of immunohistochemical markers specific to genetic alterations facilitates rapid identification of diagnostic molecular abnormalities. As such, it is currently possible for head and neck pathologists to benefit from a molecularly defined tumor classification while making diagnoses that are still based largely on histopathology and immunohistochemistry. This review covers the principal molecular alterations in sinonasal malignancies, such as alterations in DEK, AFF2, NUTM1, IDH1-2, and SWI/SNF genes in particular, that are important from a practical standpoint for diagnosis, prognosis, and prediction of response to treatment.

The prognostic role of single cell invasion and nuclear diameter in early oral tongue squamous cell carcinoma.

BACKGROUND: The clinical significance of single cell invasion and large nuclear diameter is not well documented in early-stage oral tongue squamous cell carcinoma (OTSCC). METHODS: We used hematoxylin and eosin-stained sections to evaluate the presence of single cell invasion and large nuclei in a multicenter cohort of 311 cases treated for early-stage OTSCC. RESULTS: Single cell invasion was associated in multivariable analysis with poor disease-specific survival (DSS) with a hazard ratio (HR) of 2.089 (95% CI 1.224-3.566, P = 0.007), as well as with disease-free survival (DFS) with a HR of 1.666 (95% CI 1.080-2.571, P = 0.021). Furthermore, large nuclei were associated with worse DSS (HR 2.070, 95% CI 1.216-3.523, P = 0.007) and with DFS in multivariable analysis (HR 1.645, 95% CI 1.067-2.538, P = 0.024). CONCLUSION: Single cell invasion and large nuclei can be utilized for classifying early OTSCC into risk groups.

Interpretable machine learning model for prediction of overall survival in laryngeal cancer.

Background: The mortality rates of laryngeal squamous cell carcinoma cancer (LSCC) have not significantly decreased in the last decades.Objectives: We primarily aimed to compare the predictive performance of DeepTables with the state-of-the-art machine learning (ML) algorithms (Voting ensemble, Stack ensemble, and XGBoost) to stratify patients with LSCC into chance of overall survival (OS). In addition, we complemented the developed model by providing interpretability using both global and local model-agnostic techniques.Methods: A total of 2792 patients in the Surveillance, Epidemiology, and End Results (SEER) database diagnosed with LSCC were reviewed. The global model-agnostic interpretability was examined using SHapley Additive exPlanations (SHAP) technique. Likewise, individual interpretation of the prediction was made using Local Interpretable Model Agnostic Explanations (LIME).Results: The state-of-the-art ML ensemble algorithms outperformed DeepTables. Specifically, the examined ensemble algorithms showed comparable weighted area under receiving curve of 76.9, 76.8, and 76.1 with an accuracy of 71.2%, 70.2%, and 71.8%, respectively. The global methods of interpretability (SHAP) demonstrated that the age of the patient at diagnosis, N-stage, T-stage, tumor grade, and marital status are among the prominent parameters.Conclusions: A ML model for OS prediction may serve as an ancillary tool for treatment planning of LSCC patients.

Molecular pathology in diagnosis and prognostication of head and neck tumors.

Classification of head and neck tumors has evolved in recent decades including a widespread application of molecular testing in tumors of the salivary glands, sinonasal tract, oropharynx, nasopharynx, and soft tissue. Availability of new molecular techniques allowed for the definition of multiple novel tumor types unique to head and neck sites. Moreover, the expanding spectrum of immunohistochemical markers facilitates a rapid identification of diagnostic molecular abnormalities. As such, it is currently possible for head and neck pathologists to benefit from a molecularly defined classifications, while making diagnoses that are still based largely on histopathology and immunohistochemistry. This review highlights some principal molecular alterations in head and neck neoplasms presently available to assist pathologists in the practice of diagnosis, prognostication and prediction of response to treatment.

Characterization of Immune Cell Populations of Cutaneous Neurofibromas in Neurofibromatosis 1.

Cutaneous neurofibromas (cNFs) are characteristic of neurofibromatosis 1 (NF1), yet their immune microenvironment is incompletely known. A total of 61 cNFs from 10 patients with NF1 were immunolabeled for different types of T cells and macrophages, and the cell densities were correlated with clinical characteristics. Eight cNFs and their overlying skin were analyzed for T cell receptor CDR domain sequences, and mass spectrometry of 15 cNFs and the overlying skin was performed to study immune-related processes. Intratumoral T cells were detected in all cNFs. Tumors from individuals younger than the median age of the study participants (33 years), growing tumors, and tumors smaller than the data set median showed increased T cell density. Most samples displayed intratumoral or peritumoral aggregations of CD3-positive cells. T cell receptor sequencing demonstrated that the skin and cNFs host distinct T cell populations, whereas no dominant cNF-specific T cell clones were detected. Unique T cell clones were fewer in cNFs than in skin, and mass spectrometry suggested lower expression of proteins related to T cell-mediated immunity in cNFs than in skin. CD163-positive cells, suggestive of M2 macrophages, were abundant in cNFs. Human cNFs have substantial T cell and macrophage populations that may be tumor-specific.

Prognostic Significance of Tumor-associated Stroma in Nasopharyngeal Carcinoma: A Multicenter Study.

Assessment of tumor-associated stroma has shown a reliable prognostic value in recent research. We evaluated the prognostic value of tumor-stroma ratio (TSR) in a large multicenter cohort of nasopharyngeal carcinoma (NPC). We used the conventional hematoxylin and eosin-stained slides of 115 cases of NPC to assess TSR as described in recent guidelines. The amount of tumor-associated stroma was assessed as a percentage and then tumors were classified as stroma-high (>50%) or stroma-low (≤50%). Kaplan-Meier curves, χ 2 test, and Cox regression univariable and multivariable analyses were carried out. A total of 48 (41.7%) tumors were stroma-high and 67 (58.3%) tumors were stroma-low. In the Cox regression multivariable analysis, the tumors categorized as stroma-high were associated with a worse overall survival with a hazard ratio of 2.30 (95% CI: 1.27-4.15, P =0.006) and with poor disease-specific survival (hazard ratio=1.87, 95% CI: 1.07-3.28, P =0.029). The assessment of TSR in NPC is simple and cost-effective, and it has a significant prognostic value. TSR can aid in risk stratification and clinical decision-making in NPC.

SMARCB1-deficient sinonasal adenocarcinoma: a rare variant of SWI/SNF-deficient malignancy often misclassified as high-grade non-intestinal-type sinonasal adenocarcinoma or myoepithelial carcinoma.

SMARCB1-deficient sinonasal adenocarcinoma is a rare variant of SWI/SNF-deficient malignancies with SMARCB1 loss and adenocarcinoma features. More than 200 high-grade epithelial sinonasal malignancies were retrieved. A total of 14 cases exhibited complete SMARCB1 (INI1) loss and glandular differentiation. SMARCA2 and SMARCA4 were normal, except for one case with a loss of SMARCA2. Next-generation sequencing (NGS) and/or fluorescence in situ hybridization (FISH) revealed an alteration in the SMARCB1 gene in 9/13 cases, while 2/13 were negative. Two tumors harbored SMARCB1 mutations in c.157C > T p.(Arg53Ter) and c.842G > A p.(Trp281Ter). One harbored ARID1B mutations in c.1469G > A p.(Trp490Ter) and MGA c.3724C > T p.(Arg1242Ter). Seven tumors had a SMARCB1 deletion. One carried an ESR1 mutation in c.644-2A > T, and another carried a POLE mutation in c.352_374del p.(Ser118GlyfsTer78). One case had a PAX3 mutation in c.44del p.(Gly15AlafsTer95). Histomorphology of SMARCB1-deficient adenocarcinoma was oncocytoid/rhabdoid and glandular, solid, or trabecular in 9/14 cases. Two had basaloid/blue cytoplasm and one showed focal signet ring cells. Yolk sac tumor-like differentiation with Schiller-Duval-like bodies was seen in 6/14 cases, with 2 cases showing exclusively reticular-microcystic yolk sac pattern. Follow-up of a maximum of 26 months (median 10 months) was available for 8/14 patients. Distant metastasis to the lung, liver, mediastinum, bone, and/or retroperitoneum was seen in 4/8 cases. Locoregional failure was seen in 75% of patients, with 6/8 local recurrences and 3 cervical lymph node metastases. At the last follow-up, 5 of 8 (62%) patients had died of their disease 2 to 20 months after diagnosis (median 8.2 months), and 3 were alive with the disease. The original diagnosis was usually high-grade non-intestinal-type adenocarcinoma or high-grade myoepithelial carcinoma. A correct diagnosis of these aggressive tumors could lead to improved targeted therapies with potentially better overall disease-specific survival.

Predictive value of tumor budding in head and neck squamous cell carcinoma: an update.

Head and neck squamous cell carcinoma forms an anatomically and functionally complex group of malignancies. The significant local aggressiveness and frequent regional relapses motivate ongoing research to identify more reliable and sensitive prognostic and predictive biomarkers. One emerging area of cancer biology is the evaluation of tumor budding at the advancing invasive front of various types of epithelial cancers. Recent studies suggest that tumor budding is a relatively common phenomenon in cancer progression and that it may have important prognostic implications for patients due to its potential to provide valuable insights into the biology and clinical behavior of head and neck cancer. In this review, we aim to provide information about tumor budding in head and neck squamous cell carcinoma. Thus, we hope to shed light on the complex biology of these malignancies, as well as aiding diagnostic, classification, and better characterization and thereby, looking for new avenues for improving patient outcomes.

The risk of second primary cancer after nasopharyngeal cancer: a systematic review.

PURPOSE: Second primary cancers (SPCs) after nasopharyngeal cancer (NPC) are rare, but have an impact on the follow-up of this patient population. The aim of this study is to systematically review the literature to determine the prevalence and most typical sites of SPCs after NPC. METHODS: We searched the databases of PubMed, Web of Science, and Scopus for articles on SPCs after NPC. The Preferred Reporting Items for Systematic Review and Meta-Analyses guidelines were followed. RESULTS: This review includes data on 89 168 patients with NPC from 21 articles. The mean occurrence for SPCs was 6.6% and varied from 4.9% in endemic areas to 8.7% in non-endemic areas. The most frequent locations of SPCs were oral cavity, pharynx, nose and paranasal sinuses, esophagus and lung. CONCLUSION: There is an increased risk for a SPC after NPC management, especially in non-endemic areas. However, their mean rate is lower than after other head and neck carcinomas.

Rearrangement of KMT2A Characterizes a Subset of Pediatric Parotid Mucoepidermoid Carcinomas Arising Metachronous to Acute Lymphoblastic Leukemia.

Introduction: Metachronous mucoepidermoid carcinomas (MMEC) may occur in association with childhood leukemias and lymphomas. We compared molecular abnormalities of MMEC in patients with ALL with the abnormalities found in primary mucoepidermoid carcinomas (MECs) in pediatric cases and young adults. Materials and methods: Immunohistochemical stains for p63 and SOX10, molecular alterations in MAML2 and KMT2A genes detected by FISH and/or next-generation sequencing were studied in 12 pediatric MMECs secondary to ALL and six primary MECs in pediatric patients and young adults. Follow-up information of patients in both groups was obtained. Results: KMT2A rearrangements were detected in pediatric MMECs, and they were associated with remarkable histomorphological changes, including deposits of abundant stromal collagen and intratumoral lymphoid proliferations. No KMT2A rearrangements were found in primary MECs. The prognosis of MMEC in patients with ALL, especially in KMT2A-rearranged cases, was worse than in primary MECs. Kruskal-Wallis test showed a statistically significant difference in overall survival between KMT2A-rearranged MMECs and KMT2A-intact MMECs in cases with ALL (p = 0.027). Conclusion: KMT2A-rearranged MMECs in ALL patients may have inherently more aggressive behavior, even when the histomorphology of MMEC suggests a low-grade malignancy.

Machine learning explainability in nasopharyngeal cancer survival using LIME and SHAP.

Nasopharyngeal cancer (NPC) has a unique histopathology compared with other head and neck cancers. Individual NPC patients may attain different outcomes. This study aims to build a prognostic system by combining a highly accurate machine learning model (ML) model with explainable artificial intelligence to stratify NPC patients into low and high chance of survival groups. Explainability is provided using Local Interpretable Model Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) techniques. A total of 1094 NPC patients were retrieved from the Surveillance, Epidemiology, and End Results (SEER) database for model training and internal validation. We combined five different ML algorithms to form a uniquely stacked algorithm. The predictive performance of the stacked algorithm was compared with a state-of-the-art algorithm-extreme gradient boosting (XGBoost) to stratify the NPC patients into chance of survival groups. We validated our model with temporal validation (n = 547) and geographic external validation (Helsinki University Hospital NPC cohort, n = 60). The developed stacked predictive ML model showed an accuracy of 85.9% while the XGBoost had 84.5% after the training and testing phases. This demonstrated that both XGBoost and the stacked model showed comparable performance. External geographic validation of XGBoost model showed a c-index of 0.74, accuracy of 76.7%, and area under curve of 0.76. The SHAP technique revealed that age of the patient at diagnosis, T-stage, ethnicity, M-stage, marital status, and grade were among the prominent input variables in decreasing order of significance for the overall survival of NPC patients. LIME showed the degree of reliability of the prediction made by the model. In addition, both techniques showed how each feature contributed to the prediction made by the model. LIME and SHAP techniques provided personalized protective and risk factors for each NPC patient and unraveled some novel non-linear relationships between input features and survival chance. The examined ML approach showed the ability to predict the chance of overall survival of NPC patients. This is important for effective treatment planning care and informed clinical decisions. To enhance outcome results, including survival in NPC, ML may aid in planning individualized therapy for this patient population.

Salivary Gland Secretory Carcinoma: Clinicopathologic and Genetic Characteristics of 215 Cases and Proposal for a Grading System.

Salivary gland secretory carcinoma (SC), previously mammary analog SC, is a low-grade malignancy characterized by well-defined morphology and an immunohistochemical and genetic profile identical to SC of the breast. Translocation t(12;15)(p13;q25) resulting in the ETV6 :: NTRK3 gene fusion is a characteristic feature of SC along with S100 protein and mammaglobin immunopositivity. The spectrum of genetic alterations for SC continues to evolve. The aim of this retrospective study was to collect data of salivary gland SCs and to correlate their histologic, immunohistochemical, and molecular genetic data with clinical behavior and long-term follow-up. In this large retrospective study, we aimed to establish a histologic grading scheme and scoring system. A total of 215 cases of salivary gland SCs diagnosed between 1994 and 2021 were obtained from the tumor registries of the authors. Eighty cases were originally diagnosed as something other than SC, most frequently acinic cell carcinoma. Lymph node metastases were identified in 17.1% (20/117 cases with available data), with distant metastasis in 5.1% (6/117). Disease recurrence was seen in 15% (n=17/113 cases with available data). The molecular genetic profile showed ETV6 :: NTRK3 gene fusion in 95.4%, including 1 case with a dual fusion of ETV6 :: NTRK3 and MYB :: SMR3B . Less frequent fusion transcripts included ETV6 :: RET (n=12) and VIM :: RET (n=1). A 3-tiered grading scheme using 6 pathologic parameters (prevailing architecture, pleomorphism, tumor necrosis, perineural invasion (PNI), lymphovascular invasion (LVI), and mitotic count and/or Ki-67 labeling index) was applied. Grade 1 histology was observed in 44.7% (n=96), grade 2 in 41.9% (n=90), and grade 3 in 13.5% (n=29) of cases. Compared with low-grade and intermediate-grade SC, high-grade tumors were associated with a solid architecture, more prominent hyalinization, infiltrative tumor borders, nuclear pleomorphism, presence of PNI and/or LVI, and Ki-67 proliferative index >30%. High-grade transformation, a subset of grade 2 or 3 tumors, seen in 8.8% (n=19), was defined as an abrupt transformation of conventional SC into high-grade morphology, sheet-like growth, and a tumor lacking distinctive features of SC. Both overall survival and disease-free survival (5 and 10 y) were negatively affected by tumor grade, stage, and TNM status (each P <0.0001). SC is a low-grade malignancy with predominantly solid-microcystic growth patterns, driven by a gene fusion, most commonly ETV6 :: NTRK3 . There is a low risk for local recurrence and a good overall long-term survival, with a low risk for distant metastasis but a higher risk for locoregional lymph node metastasis. The presence of tumor necrosis, hyalinization, PNI and/or LVI, and positive resection margins correlate with higher tumor grade, less favorable prognosis, and increased mortality. The statistical results allowed us to design a 3-tiered grading system for salivary SC.

Application of artificial intelligence for overall survival risk stratification in oropharyngeal carcinoma: A validation of ProgTOOL.

BACKGROUND: In recent years, there has been a surge in machine learning-based models for diagnosis and prognostication of outcomes in oncology. However, there are concerns relating to the model's reproducibility and generalizability to a separate patient cohort (i.e., external validation). OBJECTIVES: This study primarily provides a validation study for a recently introduced and publicly available machine learning (ML) web-based prognostic tool (ProgTOOL) for overall survival risk stratification of oropharyngeal squamous cell carcinoma (OPSCC). Additionally, we reviewed the published studies that have utilized ML for outcome prognostication in OPSCC to examine how many of these models were externally validated, type of external validation, characteristics of the external dataset, and diagnostic performance characteristics on the internal validation (IV) and external validation (EV) datasets were extracted and compared. METHODS: We used a total of 163 OPSCC patients obtained from the Helsinki University Hospital to externally validate the ProgTOOL for generalizability. In addition, PubMed, OvidMedline, Scopus, and Web of Science databases were systematically searched according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. RESULTS: The ProgTOOL produced a predictive performance of 86.5% balanced accuracy, Mathew's correlation coefficient of 0.78, Net Benefit (0.7) and Brier score (0.06) for overall survival stratification of OPSCC patients as either low-chance or high-chance. In addition, out of a total of 31 studies found to have used ML for the prognostication of outcomes in OPSCC, only seven (22.6%) reported a form of EV. Three studies (42.9%) each used either temporal EV or geographical EV while only one study (14.2%) used expert as a form of EV. Most of the studies reported a reduction in performance when externally validated. CONCLUSION: The performance of the model in this validation study indicates that it may be generalized, therefore, bringing recommendations of the model for clinical evaluation closer to reality. However, the number of externally validated ML-based models for OPSCC is still relatively small. This significantly limits the transfer of these models for clinical evaluation and subsequently reduces the likelihood of the use of these models in daily clinical practice. As a gold standard, we recommend the use of geographical EV and validation studies to reveal biases and overfitting of these models. These recommendations are poised to facilitate the implementation of these models in clinical practice.