AI Combines Multiple Imaging Techniques to Enhance Understanding and Prognosis of Head and Neck Cancers

Cancer researchers are leveraging artificial intelligence (AI) to combine different types of imaging data, providing deeper insights into head and neck cancers. Dr. Anant Madabhushi, executive director of the Emory Empathetic AI for Health Institute and a researcher at the Winship Cancer Institute, emphasizes the importance of integrating macroscale and microscale data to create a comprehensive portrait of tumors. This approach is crucial for understanding tumor behavior and improving risk assessments and prognoses. Four Recent Studies Virtual IHC Slides with VISTA: Researchers developed an AI platform called VISTA to transform standard H&E (hematoxylin and eosin) tissue slides into virtual immunohistochemistry (IHC) slides. These virtual IHC slides helped identify tumor-associated macrophages (TAMs), which are important for cancer prognosis. The study, published in the European Journal of Cancer, demonstrated that VISTA can detect TAMs, which are typically difficult to identify on standard H&E slides. Combining Data with Swin Transformer: The second study, published in JAMA Network Open, used a Swin Transformer to merge data from pre-treatment CT scans of primary throat cancer tumors and nearby lymph nodes. This combination of data from both the tumor and lymph nodes on CT scans was highly correlated with long-term prognosis for head and neck cancer. By analyzing these combined features, the researchers found a strong association with patient outcomes. Swin Transformer-Based Multimodal and Multi-Region Data Fusion Framework (SMuRF): Study three, published in eBioMedicine, modified the Swin Transformer into SMuRF, enabling seamless switching between 2D H&E tissue slide images and 3D radiological images. Integrating both types of images allowed the team to predict patient survival and identify those who would benefit most from chemotherapy. The model's ability to fuse different scales of data revealed subtle tumor patterns and improved prognostic accuracy. Linking Slide Images with Epigenetic Data: The fourth study, published in the European Journal of Cancer, introduced pathogenomic fingerprinting, a method that links slide images with epigenetic data. This model bridges the gap between the visual architecture of tumor cells and their genetic control, offering a more molecular understanding of the tumor. The findings contribute to a more nuanced classification of tumor aggressiveness and progression risk. Key Insights and Outcomes Madabhushi highlights that head and neck cancers, particularly oropharyngeal tumors, are complex and growing at epidemic proportions. These cancers vary significantly in shape and size, making diagnosis and treatment challenging. By applying AI to combine different imaging techniques and data types, researchers can better classify patient risks and predict outcomes. VISTA enhanced the identification of TAMs, which play a significant role in cancer prognosis. Swin Transformer and its modification into SMuRF allowed the integration of radiographic and microscopic data, revealing critical tumor patterns. Pathogenomic Fingerprinting linked visual and genetic data, advancing molecular understanding and risk assessment. Despite the promising results, Prof. Nabil Saba, co-author and the Halpern Chair in Head and Neck Cancer Research at the Winship Institute, advocates for cautious implementation. He notes that while generating large volumes of data is valuable, ensuring that this data translates effectively into clinical practice is crucial. The challenge lies in analyzing the data in the context of individual patient care to provide the best possible treatment. Saba believes that the current stage is focused on understanding the capabilities of AI in cancer research. Moving forward, the next steps involve refining these tools and validating their clinical utility. Industry insiders laud the innovative approach, recognizing the potential of AI to revolutionize cancer diagnosis and treatment. The Emory Empathetic AI for Health Institute is at the forefront of this research, aiming to develop actionable tools that can be integrated into clinical workflows. The collaborative efforts of oncologists and AI experts are paving the way for more personalized and effective cancer care. While promising, the practical application of these AI tools in clinics requires further validation and caution to ensure patient safety and efficacy. These studies represent a significant step forward in the field, demonstrating the power of AI in enhancing the precision of cancer risk assessments and treatment decisions.