Semantic segmentation of bone structures in chest X-rays including unhealthy radiographs: A robust and accurate approach

•To the best of our knowledge, this is the first work that evaluates the robustness of the bone segmentation method on severely unhealthy chest x-rays.•We proposed encoder-decoder architecture based on the combination of U-Net and Deeplab v3+ network to achieve robustness and superior performance on normal and abnormal chest x-rays.•We use pre-trained ResNet50 as a base network in the encoder and their fully connected layers have been replaced by ASPP block for improving the quality of the embedding.•We consider symmetry in down sampling and up sampling steps to concatenate both low-level and high-level feature maps ensuring better embedding of both the edges and detail information.•At each level, the up-sampled decoder features are concatenated with the encoder features at a similar level and further passes to fine-tuning block for better segmentation.