Using recent advances in machine learning, a team of scientists led by Saeed Hassanpour, PhD, assistant professor of biomedical data science at Dartmouth Geisel School of Medicine, developed a deep neural network to classify different types of lung adenocarcinoma on histopathology slides, and found that the model performed on par with three practicing pathologists.
Currently, lung adenocarcinoma requires pathologist’s visual examination of lobectomy slides to determine the tumor patterns and subtypes. This classification has an important role in prognosis and determination of treatment for lung cancer, but it is a difficult and subjective task, according to Hassanpour.
Using recent advances in machine learning, the scientists, led by Hassanpour, developed a deep neural network to classify different types of lung adenocarcinoma on histopathology slides, and found that the model performed on par with three practicing pathologists.
Visualization on sample whole-slide images of the lung cancer histologic patterns identified by pathologists compared to those detected by a new machine learning model developed by researchers at Dartmouth’s Norris Cotton Cancer Center. The team rendered the image by overlaying color-coded dots on patches based on decisions generated by their computer model. A subjective qualitative assessment by pathologist annotators confirmed that the patterns detected on each slide are on target. [Hassanpour Lab, Dartmouth’s Norris Cotton Cancer Center][/caption]
“Our study demonstrates that machine learning can achieve high performance on a challenging image classification task and has the potential to be an asset to lung cancer management,” says Hassanpour. “Clinical implementation of our system would be able to assist pathologists for accurate classification of lung cancer subtypes, which is critical for prognosis and treatment.”
The team’s conclusions (“Pathologist-level classification of histologic patterns on resected lung adenocarcinoma slides with deep neural networks”) are published in Scientific Reports. Recognizing that the approach is potentially applicable to other histopathology image analysis tasks, Hassanpour’s team made their code publicly available to promote new research and collaborations in this domain.
In addition to testing the deep learning model in a clinical setting to validate its ability to improve lung cancer classification, the team plans to apply the method to other challenging histopathology image analysis tasks in breast, esophageal, and colorectal cancer. “If validated through clinical trials, our neural network model can potentially be implemented in clinical practice to assist pathologists,” said Hassanpour. “Our machine learning method is also fast and can process a slide in less than one minute, so it could help triage patients before examination by physicians and potentially greatly assist pathologists in the visual examination of slides.”