A deep neural network crafted by research specialists at Dartmouth’s Norris Cotton Cancer Center identified different types of lung adenocarcinoma as well as practicing pathologists in a recent study, according to work published March 4 in Scientific Reports.
“Classification of histologic patterns in lung adenocarcinoma is critical for determining tumor grade and treatment for patients,” first author Jason W. Wei and colleagues wrote in the journal. “However, this task is often challenging due to the heterogeneous nature of lung adenocarcinoma and the subjective criteria for evaluation.”
Adenocarcinoma is the most common type of lung cancer, the authors said, and identifying tumor patterns and subtypes typically requires the expertise of a trained pathologist. With advances in machine learning, though, Wei et al. were able to develop a deep learning model that automatically classified the histologic patterns of lung adenocarcinoma on surgical resection slides.
The team’s model uses a convolutional neural network to identify regions of neoplastic cells, then aggregates those classifications to infer minor and predominant histologic patterns They tested the approach against three pathologists’ professional opinions on an independent set of 143 whole-slide images.
The researchers said their model achieved a kappa score of 0.525 and an agreement of 66.6 percent with the pathologists for classifying predominant patterns—marginally higher than the inter-pathologist kappa score of 0.485 and agreement of 62.7 percent.
“Our model slightly edged out the pathologists on these two metrics, possibly because computing tumor areas by counting the number of patches is more precise than unaided estimations of tumor area by the naked eye,” Wei and co-authors wrote.
The team said 39.5 percent of disagreements in predominant subtype classification were between the acinar and lepidic subtypes, which makes sense since those two patterns often appear together and can be difficult to distinguish. Detecting minor patterns was more difficult for both the AI and pathologists.
Wei and colleagues said they next plan to apply their model to other histopathology image analysis in breast, esophageal and colorectal cancer. They’ve also made the code for their model publicly available here.
“If confirmed in clinical practice, our model can assist pathologists in improving classification of lung adenocarcinoma patterns by automatically pre-screening and highlighting cancerous regions prior to review,” the authors wrote. “Our approach can be generalized to any whole-slide image classification task.”
After graduating from Indiana University-Bloomington with a bachelor’s in journalism, Anicka joined TriMed’s Chicago team in 2017 covering cardiology. Close to her heart is long-form journalism, Pilot G-2 pens, dark chocolate and her dog Harper Lee.