Automatic Lung Field Segmentation using Robust Deep Learning Criteria

DOI:10.21742/ijhit.2653-309X.2021.1.1.06

AUTHORS

Mohammed Ryiad Al-Eiadeh,Al Yarmouk University, Jordan

ABSTRACT

The chest X-Ray image is one of the common image types that describe medical imaging for medical purposes, its more common than other medical image types like MRI, CT scan, and PET scan. Huge medical tests that depend on the CXR-images are difficult to deal with X-Ray images on both radiologists and medical practitioners. the organ partitioning is a step to obtain an efficient computer-aided detection and how the computer vision can be used to provide good functionality in medical aided diagnosis systems depending on the deep learning. Lung field segmentation plays an important role in medical diagnosis systems like a task to classify whether the image belongs to a normal or abnormal person based on the shape of the segmented or predicted lung field. We proposed a model that combined the DeepLabv3 plus with ResNet-18 to build a segmentation network that performs decoding and encoding phases to extract the spatial features of a particular area and output the segmented one, respectively. The model is evaluated on a public dataset Montgomery County set - Chest X-ray Database, multiple experiments were conducted by dividing the data into 60% for training and the rest 40% for testing, the experiment was repeated 10 times as for each time data divisors were randomly selected to support our model performance. Our model Performance measurements achieved Dice=96.9 % and Jaccard=94.1% which shown the robustness of our proposed approach.