AI Outperforms Expert Pathologists in Predicting Lung Cancer Spread
Posted on 12 Mar 2024
For years, the medical community has been struggling with the challenge of predicting which lung cancer patients are most likely to experience metastasis. This knowledge is crucial for treating early-stage non-small cell lung cancer (NSCLC) patients, as it influences whether they should undergo aggressive treatments like chemotherapy or radiation after lung surgery. Over half of stage I–III NSCLC patients eventually face brain metastasis, but for many others, such intensive treatments are unnecessary. Now, researchers have found that artificial intelligence (AI) could be a promising tool in aiding physicians with these critical decisions.
A groundbreaking pilot study conducted by Caltech (Pasadena, CA, USA) and Washington University School of Medicine in St. Louis (WUSTL, St. Louis, Mo, USA) revealed AI's capability to outperform expert pathologists in predicting the likelihood of cancer metastasis in NSCLC patients. The study involved training a deep-learning network, a sophisticated type of AI program, using hundreds of thousands of image tiles derived from biopsy images of 118 NSCLC patients. These images are typically reviewed by pathologists for cell abnormalities indicating cancer progression. The AI was tested with 40 additional biopsy images to assess its ability to predict brain metastases, demonstrating a striking 87% accuracy, surpassing the 57% accuracy rate of four expert pathologists.
.jpg "Image: AI outperformed expert pathologists in predicting which lung cancer cases are likely to metastasize (Photo courtesy of Shutterstock/Kateryna Kon)")
Notably, the AI's predictions were even more accurate for the earliest-stage NSCLC patients (stage I) and were based on standard microscopic slides. The researchers believe that incorporating more data, such as disease severity and biomarkers, could enhance the AI's predictive capabilities. However, the researchers caution that this is just an initial step, and a larger study is necessary to validate these findings. Interestingly, the AI doesn't explicitly reveal the factors influencing its predictions, prompting ongoing research to decode the complex tumor cell features and their environment it might be analyzing. Going forward, Caltech scientists aim to develop improved instrumentation and procedures for collecting uniform, high-quality biopsy images, which could further refine the accuracy of AI predictions in cancer treatment.
"Overtreatment of cancer patients is a big problem," said Changhuei Yang, the Thomas G. Myers Professor of Electrical Engineering, Bioengineering, and Medical Engineering at Caltech. "Our pilot study indicates that AI may be very good at telling us in particular which patients are very unlikely to develop brain cancer metastasis."
"Our study is an indication that AI methods may be able to make meaningful predictions that are specific and sensitive enough to impact patient management," added Richard Cote, head of the Department of Pathology & Immunology at WUSTL.
