kaledhoshme123 / probabilistic-u-net-segmentation-ambiguous-images-

People with pulmonary disease often have a high opacity, which makes segmentation of the lung from chest X-rays more difficult. In this study, I propose a methodology to improve the performance of the U-NET structure so that it is able to extract the features and spatial characteristics of the X-ray images of the chest region.

Probabilistic U-Net Segmentation of Ambiguous Images:

• People with pulmonary disease often have a high opacity, which makes segmentation of the lung from chest X-rays more difficult.
• In this study, I propose a methodology to improve the performance of the U-NET structure so that it is able to extract the features and spatial characteristics of the X-ray images of the chest region, in addition to the ability to find a link between the left and right parts of the chest region, which helps the neural network to locate the lungs within the X-ray images ( segmentation).
• Since the variational autoencoder is used for generation, we can use it to extract the most important information in the x-ray images, such as what we mentioned, the relationship between the left and right parts of the chest region.
• Therefore, I proposed integrating the U-NET structure with the variational autoencoder structure in order to reach a neural network structure capable of segmentation for people with lung diseases.
• To ensure the accuracy of the study, the neural network was trained on medical chest x-rays of people with lung opacity, and then the model was tested on chest x-rays of people with COVID-19.
• The results were as follows: The model was able to train it on chest x-rays of people with lung opacity: accuracy: 0.9829 - precision: 0.9939 - recall: 0.9606. After completing the training, the model was tested on chest X-rays of people infected with COVID-19, and the results obtained were accuracy: 0.9789 - precision: 0.9902 - recall: 0.9650.

The previous image shows several samples of X-ray images of a healthy person, a person with medium opacity, and a person with high opacity.

Evaluation of the performance of the model:

Evaluation of the performance of the model on x-rays of the chest area of people with lung_opacity:

Evaluation of the performance of the model on x-rays of the chest area of people with covid:

Compare original masks with those predicted by the model:

Samples of Lung_opacity Images:

Samples of COVID-19 Images: