Advancing Education, Research, and Quality of Care for the Head and Neck oncology patient.
Introduction: AI tools like ChatGPT are increasingly used in healthcare, offering clinical decision support. However, cloud-based systems raise concerns about patient privacy and data security. This study presents a locally deployed AI model for clinical decision support in head and neck oncology. The AI operates entirely offline, ensuring confidentiality while integrating National Comprehensive Cancer Network (NCCN) guidelines to provide accurate, guideline-based recommendations without internet access.
Methods: This proof-of-concept study evaluated a locally deployed AI chatbot using the Llama3-ChatQA model. The system ran on local hardware, eliminating the need for cloud-based servers and ensuring data privacy.
NCCN guidelines were integrated into a large language model. First it was converted into text, divided into manageable chunks, and encoded as vectors using the Universal Sentence Encoder (USE). These vectors were stored in a database for efficient retrieval. User queries were similarly processed, matched to relevant guideline sections, and answered accordingly. Various clinical scenarios were tested to assess the accuracy of the recommendations.
Results: The AI provided accurate, reliable responses based on NCCN guidelines. Key examples include:
HPV-positive oropharyngeal squamous cell carcinoma diagnosis: The AI recommended p16 immunohistochemistry as a primary diagnostic tool.
Nasopharyngeal carcinoma staging: MRI and CT with contrast were recommended, with PET-CT for metastasis detection.
Stage III laryngeal cancer treatment: Options included surgery, radiation, or chemotherapy, tailored to patient specifics.
Systemic therapies for metastatic head and neck squamous cell carcinoma: PD-L1 testing guides treatment. Pembrolizumab or nivolumab with chemotherapy was recommended for PD-L1-positive tumors, and platinum-based chemotherapy for others.
Trismus management post-radiation therapy: The AI suggested physical therapy and pharmacotherapy, such as cyclosporine, with botulinum toxin injections for severe cases.
Immunotherapy for recurrent head and neck cancer: PD-L1 expression guides immunotherapy use, with combination therapy to enhance outcomes.
Conclusion: This is the first of its kind in the Otolaryngology literature which demonstrates that a locally deployed AI model can provide clinical decision support without internet connection. By integrating NCCN guidelines and running offline on a desktop or laptop, the system ensures data security and delivers accurate, guideline-based recommendations. This approach addresses key concerns in healthcare AI deployment and offers a secure solution for sensitive areas like head and neck oncology providing helpful support to patients and providers.