Machine Learning Assessment of Pathologic Response in Lung Cancer Resections After Neoadjuvant Therapy-IASLC MPR
Sanja Dacic1, Daniel Shenker2, Mary Redman3
1Department of Pathology, Yale School of Medicine, New Haven, Connecticut.
Introduction:
Machine learning algorithms may improve the efficiency and accuracy of pathologic response (PR) assessment in surgically resected lung cancers after neoadjuvant therapy. The aim of this study was to develop digital models for quantifying tumor bed (TB) area and residual viable tumor (VT) and to compare these results to previously published assessments of PR by pathologists from the International Association for the Study of Lung Cancer reproducibility study.
Methods:
Manual pathologist annotations (N = 15,564) of regions including TB and VT were used to train a convolutional neural network model (digital artificial intelligence [AI]) and a convex hull algorithm (CHA). PR was determined by the percentage of VT in the TB area. The pathologists determined the average PR (APR) across slides (unweighted), which was compared with the weighted average for digital AI and CHA. The concordance between pathologist APR, digital AI, and CHA was calculated and correlated with outcomes.
Results:
There was a strong correlation between approaches: APR versus digital AI (0.97), APR versus CHA (0.97), and digital AI versus CHA (0.99). Digital AI and CHA demonstrated 100% agreement for MPR. The κ concordance for MPR was 0.82 (95% confidence interval [CI]: 0.69, 0.96) for APR versus digital AI/CHA with six discordant cases. The concordance was higher for squamous cell carcinoma (κ = 0.92, 95% CI: 0.76, 1.0) than for nonsquamous carcinoma (κ = 0.77, 95% CI: 0.59, 0.96). APR and digital AI demonstrated similar relapse-free survival and overall survival.
Conclusion:
The overall high level of agreement supports the utility of machine learning approaches for evaluation of PR in patients with NSCLC.