We present a novel graph-based approach for labeling the anatomical branches of a given airway tree segmentation. The proposed method formulates airway labeling as a branch classification problem in the airway tree graph, where branch features are extracted using convolutional neural networks (CNN) and enriched using graph neural networks. Our graph neural network is structure-aware by having each node aggregate information from its local neighbors and position-aware by encoding node positions in the graph. We published our source code. The algorithm is also publicly available as an algorithm served on the grand-challenge website.

I propose a weakly-supervised segmentation method based on dense regression activation maps (dRAMs). Most weakly-supervised segmentation approaches exploit class activation maps (CAMs) to localize objects. However, because CAMs were trained for classification, they do not align precisely with the object segmentations. Instead, my method produce high-resolution activation maps using dense features from a segmentation network that was trained to estimate a per-lobe lesion percentage. In this way, the network can exploit knowledge regarding the required lesion volume. We evaluated our algorithm on 90 subjects. Results show our method achieved 70.2% Dice coefficient, substantially outperforming the CAM-based baseline at 48.6%.

I propose a deep learning clustering method that exploits dense features from a segmentation network for emphysema subtyping from computed tomography (CT) scans. This approach provides model interpretability via dense clustering activation maps (dCAMs). On the evaluation dataset with 500 subjects, the method achieved a 43% unsupervised clustering accuracy, 0.54 in silhouette coefficient, and 0.55 in David-Bouldin scores.

Segmenting pulmonary lobes in Chest CT is important for regional quantitative analysis of many lung diseases such as Chronic Obstructive Pulmonary Diseases (COPD). I designed a method based on deep neural networks to segment lobes from the Chest CT images automatically. My method currently ranks 3 out of 818 teams worldwide in Lola11 Grand-challenge (25th in overall Lung segmentation ranking including submissions with human annotations). I also made this algorithm publically available as an algorithm served at the Grand-challenge website. This algorithm is being used by 167 users worldwide.

I developed algorithms to automatically detect pulmonary nodules in Chest CT and assess of the malignancy risk of found nodules. My method currently ranks the 2nd out of 1000+ teams worldwide in Luna16 Grand-challenge.