Advancing Education, Research, and Quality of Care for the Head and Neck oncology patient.
Background: In recent years, there have been various efforts to integrate artificial intelligence (AI) into the surgical setting, particularly in otolaryngology. Recognizing the vast opportunities that AI offers in enhancing patient outcomes, we developed a tailored AI model designed to predict the prognosis of individual patients with salivary gland malignancies. This model personalizes prognostic predictions by analyzing a range of patient-specific factors, including demographic details (such as age, race, gender, and socioeconomic status), radiographic findings, histopathology, and treatment history. The treatment information encompasses various approaches, detailing whether surgery was performed, the type of surgery, and any radiation or chemotherapy administered.
Methods: An AI model for data analysis, harmonization, and treatment optimization was developed, trained, and tested using the SEER dataset of 20174 total patients with salivary gland malignancy. This deep attention model prioritized the individual patient characteristic to learn the survival probabilities of 1, 2, 5, and 10 years to predict prognosis as well as the expected survival of each patient. A prognosis of survival was then calculated that is able to adapt depending on the individual characteristics and selected treatment of each patient. We obtained the accuracy, precision, recall, F1 score, and area under the curve (AUC) of the performance of each model.
Results: The results for the 1, 2, 5, and 10-year survival probabilities from various models are as follows. The Deep Learning developed model showed an AUC of 0.83, 0.79, 0.75, and 0.73 for the 1, 2, 5, and 10 year survival prediction. In comparison, the XGBoost model achieved an AUC of 0.79, 0.72, 0.52, and 0.50 for the 1, 2, 5, and 10 year survival prediction. Finally, the MLPClassifier model showed an AUC of 0.71, 0.67, 0.57, 0.52 for the 1, 2, 5, and 10 year survival prediction.
Conclusions: In this study, we developed an AI model that is able to optimize and personalize prognosis prediction for salivary gland cancer patients using a personalized approach for each patient. To date, there has not been a developed AI model that is able to predict patient survival and prognosis through integrating and adapting patient specific factors and treatment. We also show that our model is able to more accurately predict long term prognosis (5-year and 10-year survival predictions) with higher accuracy compared to standard prognosis prediction methods. Our aim in this paper is to offer this AI tool to physicians to assist otolaryngologists and patients in determining a personalized prognosis based on their individual characteristics and selected treatment.