Intelligent Three-Dimensional Optical Coherence Tomography for Lung Tumor Classification

Accurate intraoperative grading of lung tumors is challenging with frozen-section pathology. We present a intelligent three-dimensional optical coherence tomography framework for automated classification. Volumetric datasets from sequential B-scans were analyzed via slice-based multiple instance learning and a 3D convolutional network. Binary classification of invasive adenocarcinoma versus normal tissue achieved 0.91 accuracy in both models, with the volumetric model showing more stable feature separation. Multi-class classification remained limited by structural similarity and tumor heterogeneity.