April Salama’s War on Metastatic Melanoma
Published
Written By
Julie Poucher Harbin
Senior Writer, Duke Cancer Institute
Related News
New AI Methods Advance Diagnosis After Brain Metastasis Treatment
New research from a team of experts at Duke University School of Medicine highlights the use of artificial intelligence (AI) and computational methods to differentiate between local recurrence and radionecrosis in brain metastasis patients following stereotactic radiosurgery (SRS).Members of the Duke Center for Brain and Spine Metastasis and graduate students in the Duke Department of Radiation Oncology – including Jingtong Zhao, MS – led this research.Brain metastases are the most common type of brain tumor in adults. Among those treated with SRS, 10 to 20 percent develop radiographic changes on follow-up MRI scans. These changes can represent either local recurrence or radionecrosis, which are difficult to distinguish using imaging alone due to their similar appearance.Differentiating between local recurrence and radionecrosis using current imaging techniques, such as perfusion MRI and diffusion-weighted MRI, remains challenging due to their limitations. Zhao said she and her team identified a great need for a non-invasive clinical tool to improve diagnostic accuracy.“We believed there was an opportunity to use computational methods to extract image features from post-SRS MRI scans and develop an AI model to make decisions,” Zhao said.To address these challenges, Zhao and the team proposed a deep learning-based approach that leverages radiomics to extract image features from post-SRS MRI scans. This method involves using a deep neural network to predict whether a patient will develop radionecrosis or local recurrence.While radiomics offers valuable insights, traditional radiomics methods can struggle with imbalanced datasets and are sensitive to technical variations in imaging. Deep learning, by contrast, can automatically learn complex patterns from imaging and non-imaging data without relying on manually crafted features, making it a powerful tool for capturing nuanced differences between radionecrosis and tumor recurrence.A key challenge with deep learning models is that they often function as “black boxes,” offering little explanation for how decisions are made. Physicians may be hesitant to trust AI without understanding its logic, and Zhao emphasized the need for transparency in this decision making.The current model enhances explainability by tracking how different categories of features, such as imaging, clinical, and genomic data, contribute to the model’s predictions over time. Next steps for the team’s research will focus on modeling the dynamics of individual features to show how they contribute to the AI’s overall thought process.“Using these AI methods can help improve diagnostic accuracy,” Zhao said. “This approach has the potential to enhance patient outcomes and pave the way for future research in this field.”
Related News
New AI Methods Advance Diagnosis After Brain Metastasis Treatment
New research from a team of experts at Duke University School of Medicine highlights the use of artificial intelligence (AI) and computational methods to differentiate between local recurrence and radionecrosis in brain metastasis patients following stereotactic radiosurgery (SRS).Members of the Duke Center for Brain and Spine Metastasis and graduate students in the Duke Department of Radiation Oncology – including Jingtong Zhao, MS – led this research.Brain metastases are the most common type of brain tumor in adults. Among those treated with SRS, 10 to 20 percent develop radiographic changes on follow-up MRI scans. These changes can represent either local recurrence or radionecrosis, which are difficult to distinguish using imaging alone due to their similar appearance.Differentiating between local recurrence and radionecrosis using current imaging techniques, such as perfusion MRI and diffusion-weighted MRI, remains challenging due to their limitations. Zhao said she and her team identified a great need for a non-invasive clinical tool to improve diagnostic accuracy.“We believed there was an opportunity to use computational methods to extract image features from post-SRS MRI scans and develop an AI model to make decisions,” Zhao said.To address these challenges, Zhao and the team proposed a deep learning-based approach that leverages radiomics to extract image features from post-SRS MRI scans. This method involves using a deep neural network to predict whether a patient will develop radionecrosis or local recurrence.While radiomics offers valuable insights, traditional radiomics methods can struggle with imbalanced datasets and are sensitive to technical variations in imaging. Deep learning, by contrast, can automatically learn complex patterns from imaging and non-imaging data without relying on manually crafted features, making it a powerful tool for capturing nuanced differences between radionecrosis and tumor recurrence.A key challenge with deep learning models is that they often function as “black boxes,” offering little explanation for how decisions are made. Physicians may be hesitant to trust AI without understanding its logic, and Zhao emphasized the need for transparency in this decision making.The current model enhances explainability by tracking how different categories of features, such as imaging, clinical, and genomic data, contribute to the model’s predictions over time. Next steps for the team’s research will focus on modeling the dynamics of individual features to show how they contribute to the AI’s overall thought process.“Using these AI methods can help improve diagnostic accuracy,” Zhao said. “This approach has the potential to enhance patient outcomes and pave the way for future research in this field.”