New findings by researchers at Yale Cancer Center (YCC) show deep learning algorithms can outperforms radiologists in finding microscopic infiltrations of head and neck cancer in lymph nodes. The study, published today in the Journal of Clinical Oncology (JCO), suggests this machine learning approach or artificial intelligence (AI) could help radiologists correctly diagnose the condition known an extranodal extension (ENE). ENE is a condition where a lymph node has been invaded with cancer and the disease has extended into the surrounding tissue. Traditionally, ENE is found after an extensive surgery and necessitates the use of additional chemotherapy. If ENE can be identified preoperatively, patients could avoid having unnecessary surgery and receive treatment with just chemotherapy and radiation therapy alone.
According to the study’s senior investigator, Sanjay Aneja, M.D., assistant professor of Therapeutic Radiology at YCC and Smilow Cancer Hospital, diagnosing ENE preoperatively has been difficult, but could be very valuable in saving patients excessive surgical treatments. YCC researchers developed the deep learning technique and initially tested it in their own patient population, and in this study, broadened their examination to multiple and diverse datasets. The deep learning algorithm designed by Dr. Aneja’s Lab uses neural networks to analyze individual pixels of CT images. Researchers found performance of the deep learning analysis outperformed board-certified radiologists who specialize in head and neck cancers.
“Further testing is needed before deep learning is routinely used in clinical practice,” said Aneja. “But we feel this tool is just one example of how deep learning can help clinicians provide more personalized treatment for patients.” Aneja’s Lab is studying the utility of their deep learning technique across multiple other cancer types including lung cancer and brain tumors.