Attention UW-Net: A fully connected model for automatic segmentation and annotation of chest X-ray

Automatic segmentation and annotation of medical image plays a critical role in scientific research and the medical care community. Automatic segmentation and annotation not only increase the efficiency of clinical workflow, but also prevent overburdening of radiologists. The objective of this work is to improve the accuracy and give a probabilistic map for automatic annotation from small data set to reduce the use of tedious and prone to error manual annotations from chest X-rays.

In this paper, we have proposed an attention UW-Net, which introduces an intermediate layer acting as a bridge between the encoder and decoder pathways. The intermediate layer is a series of fully connected convolutional layers generated from the upsampling of the final encoder layer connected to the corresponding up sampled and down sampled blocks via skip-connections. The intermediate layer is further connected to the decoder pathway using a downsampling layer.

The proposed attention UW-Net is giving a very good performance, achieving an average F1-score of 95.7%, 80.9%, 81.0% and 77.6% for lung (large), heart (medium), trachea (small), and collarbone (small) object segmentations, respectively. The attention UW-Net outperforms not only in comparison to U-Net and its variations but also with respect to other standard recent automatic and semi-automatic segmentation/annotation models. An ablation study was also performed to find the best suited high-performing architecture.

The uniformity in prediction accuracy of segmentation masks for all kinds of segmentation masks (large, medium, and small lesions) makes this model best for automatic annotation of organs.