Automated Chest Cavity and Lung Segmentation for Temporal Tracking of Lung Pathologies on Chest X-Ray

Poster Presentation at the European Congress of Radiology, Vienna, 2019


Objectively evaluating progression and regression of lung pathologies on frontal Chest X-Ray (CXR) in this era of Artificial Intelligence algorithms remains a challenge. We present a novel Deep Learning (DL) based approach to segment the chest cavity on a CXR for this purpose.

Methods and Materials

A 97-layer deep UNET network was trained on 3,056 CXRs (from NIH-ChestXRay14) with each image having manually drawn (by 3 practising radiologists, supervised by a sub-specialist radiologist) masks of the Chest Cavity (CC) and bilateral ‘Lucent Lung Cavities’ (LLC), defined as the lung parenchyma as observed on CXR which includes pneumothorax, focal consolidation and nodules, but excludes pleural effusion. The network was tested on an independent dataset of 476 CXRs and Dice Score Coefficients (DSC) were calculated.


The network achieved an average DSC of 95.5%, 95.3% and 97% when comparing the predicted and ground-truth masks for left LLC, right LLC and CC respectively.


Automated segmentation of the CC and LLCs can help temporally track the extent of pathology on a CXR. Additionally, an LLC: CC ratio trend can act as a surrogate marker for a patient’s respiratory reserve in conditions like pleural effusion. More research is required to establish this claim.