A Multifaceted Radiomics Model to Predict Cervical Lymph Node Metastasis for Involved Nodal Radiation Therapy PROJECT SUMMARY The majority of disease sites treated with radiation therapy (RT) no longer receive elective/prophylactic RT to clinically-negative areas, including lung, pancreas, and lymphoma. These disease sites now employ involved nodal radiotherapy (INRT), focusing on involved lymphadenopathy. However, in head and neck cancer (HNC), we still target the same lymph node regions as conventional 2D radiotherapy, despite our ability to tailor the radiotherapy volume and dose to specific areas using intensity modulated radiation therapy (IMRT). This approach leads to excessive acute and long-term toxicities for HNC patients after RT. Therefore, INRT is highly desirable for HNC. In INRT, one particular challenge during gross tumor volume (GTV) and clinical target volume (CTV) delineation is the identification of malignant lymphadenopathy. While some lymph nodes (LNs) are obviously malignant based on standard imaging modalities, there is often uncertainty about whether a LN is malignant and requires targeting. Treating benign nodes as malignant may cause a significantly higher risk of late complications, such as xerostomia and dysphagia. On the other hand, missing occult lymphadenopathy will lead to regional recurrence. The goal of this project is to develop, optimize, and test a multifaceted predictive model with both high sensitivity and specificity for LN metastasis classification to maximize the efficacy and minimize the toxicity of INRT for HNC. The proposed multifaced model presents a flexible framework and considers multiple aspects of a predictive model, including: 1) Evaluation criteria used in model training (multi- objective); 2) Different sources of information (multi-modality); and 3) Classifiers used for model construction (multi-classifier). By designing a multi-objective function, we will consider sensitivity and specificity simultaneously during model training and optimization. Instead of blindly combining features extracted from different modalities and empirically choosing one preferred classifier, the information extracted by modality- specific classifiers will be combined optimally through a reliable classifier fusion (RCF) strategy. We will develop a prospective registry database to train the multi-classifier, multi-objective and multi-modality (MCOM) model through prospectively collecting clinical characteristics and images of HNC patients who will undergo surgery at UTSW with pathology-confirmed LN metastasis status. The model will be validated on an independent UTSW patient cohort and patients who underwent outside imaging but operated at UTSW. The specific aims of the project are: 1) Develop and validate a multi-classifier, multi-objective and multi-modality (MCOM) LN metastasis prediction model for HNC patients. 2) Conduct a randomized phase II clinical trial to evaluate the efficacy and utility of INRT versus conv...