Site Editor
Gregory J. Riely, MD, PhD
Gregory J. Riely, MD, PhD
Posted: Friday, April 5, 2024
A recent article published in The Journal of Pathology highlighted the use of a deep learning algorithm to predict the eventual development of brain metastases in patients with non–small cell lung cancer (NSCLC). Richard J. Cote, MD, of Washington University School of Medicine, St. Louis, and colleagues used images from hematoxylin and eosin (H&E)-stained tumor tissue to train the digital learning network to predict metastatic progression in brain tissue within 5 years of initial diagnosis. Their findings revealed that the algorithm was able to predict the eventual development of brain metastases with 87% accuracy.
“There are no predictive tools available to help physicians when treating patients with lung cancer,” stated Dr. Cote in an institutional press release by Marta Wegorzewska. “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.”
A total of 154 patients with NSCLC were included in this study; 113 had stage I disease, and 41 had higher-stage disease. One representative block of tumor tissue was used to create a fresh H&E slide for each patient; slides were then scanned at 40x magnification. The deep learning model was based on the ResNet-18 convolutional neural network, pretrained on the ImageNet data set. Three rounds of cross-validation training were performed for each experiment. Findings were also compared with a blinded review by four pathology experts.
Findings revealed that the deep learning–based algorithm distinguished the eventual development of brain metastases with an accuracy of 87% (P < .0001) compared with an average of 57.3% by the four pathologists. Additionally, the algorithm was particularly successful in predicting brain metastases in patients with stage I disease.
Disclosure: The study authors reported no conflicts of interest.
JNCCN Spotlights
Resources
For your Patients
