Aims and objectives
Objectively evaluating progression and regression of lung pathologies on frontal Chest X-Ray (CXR) is a challenge. Universally, the temporal evolution of most pathological conditions on chest x-ray is still subjective, which might be acceptable for directional prediction but not reliable for quantification of the change. Here we present a novel Deep Learning (DL) based approach to automatically segment the chest cavity and lungs on a CXR for this purpose. This method will be useful in the evaluation of longitudinal change in relative proportions between the aerated lung and pathological conditions.
Methods and materials
The dataset involved 3531 XRAY images (1024x1024), which were randomly separated into (2714) training, (342) validation, and (475) test sets, such that each patient had images in one set only. The chest-cavity and lung masks were available for all cases.
Model training: The XRAY images were resampled to 255x255 resolution and min-max normalization was applied to the (8-bit, grayscale) intensity. The model architecture is UNET. (The contracting part of the network involves concatenated blocks of convolution, batch-normalization and dropout, the result of that is down-sampled by a factor of 3. The expanding part involves concatenated blocks of transpose convolution, batch-normalization and dropout, the result of that is up-sampled by a factor of 3.) The model has 120 layers and 2,413,762 trainable parameters. The model was trained for 500 epochs (with batch size equal to 10). After all augmented (random shift, rotation and scaling) training samples were processed in an epoch, the model was evaluated on the validation samples and its parameters were saved if the accuracy (DICE score) was improved. Separate models were trained for chest-cavity, left- and right lung.
Results
The chest cavity segmentation closely follows the left and right lung segmentation demonstrating repeatability in both the cases (Fig. 1 and Fig. 2). With the advancements in deep learning, pneumothorax detection and segmentation algorithms are emerging within the market. The following experiment simulates what could be possible in the future with our lung segmentation algorithms paired with pneumothorax segmentation, by using manually segmented PTX regions as a substitute for the demonstration. The intersection of the ground truth PTX masks and the right/left lung segmentation was used to determine the outer boundary of the PTX region, as to represent PTX regions within the automatic lung segmentation area.
Conclusion
Automated segmentation of the Chest Cavity and Lucent lung cavities 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.
Link to complete publication here: https://scholar.google.com/citations?view_op=view_citation&hl=en&user=dqpMNRUAAAAJ&cstart=20&pagesize=80&citation_for_view=dqpMNRUAAAAJ:ufrVoPGSRksC
