Deep learning-based techniques for the automatic segmentation of organs in thoracic computed tomography images: A Comparative study

Medical images have become the important part in medical diagnosis and treatment. These images play a significant role in medical field because the doctors are highly interested in exploring the anatomy of the human body. The medical images captured with various modalities like (PET, CT, SPECT, MRI, etc.) have different variability based on the intensity level. The segmentation of organs in medical images is the most crucial image-related application. Organs segmentation in the medical images help the doctors in planning the treatment in lesser time and with higher efficiency. Results of manual segmentation vary from experts to experts and it is very time taking task. Automatic segmentation is the solution to the problem as it gives precise results. Several techniques had been addressed in the literature for the segmentation of thoracic organs (heart, aorta, trachea, esophagus) automatically in the medical images. Out of those deep learning-based techniques outperformed in the automatic segmentation of organs by giving precise accuracy. Using deep learning models for segmentation purposes improve the segmentation results in various clinical applications. In this paper various deep learning-based techniques for automatic segmentation had been discussed. Also, the three authors' results are compared based on the parameters such as Dice Coefficient (DC) and Hausdorff Metric (HM).