LAUSR

Abstract Background and objectives Deep Learning (DL) and Machine Learning (ML) have brought several breakthroughs to biomedical image analysis by making available more consistent and robust tools for the identification, classification, reconstruction, denoising, quantification, and segmentation of patterns in biomedical images. Recently, some applications of DL and ML in Computed Tomography (CT) scans for low dose optimization were developed. Nowadays, DL algorithms are used in CT to perform replacement of missing data (processing technique) such as low dose to high dose, sparse view to full view, low resolution to high resolution, and limited angle to full angle. Thus, DL comes with a new vision to process biomedical data imagery from CT scan. It becomes important to develop architectures and/or methods based on DL algorithms for minimizing radiation during a CT scan exam thanks to reconstruction and processing techniques. Methods This paper describes DL for CT scan low dose optimization, shows examples described in the literature, briefly discusses new methods used in CT scan image processing, and offers conclusions. We based our study on the literature and proposed a pipeline for low dose CT scan image reconstruction. Our proposed pipeline relies on DL and the Spark Framework using MapReduce programming. We discuss our proposed pipeline with those proposed in the literature to conclude the efficiency and importance. Results An architecture for low dose optimization using CT imagery is suggested. We used the Spark Framework to design the architecture. The proposed architecture relies on DL, and permits us to develop efficient and appropriate methods to process dose optimization with CT scan imagery. The real implementation of our pipeline for image denoising shows that we can reduce the radiation dose, and use our proposed pipeline to improve the quality of the captured image. Conclusion The proposed architecture based on DL is complete and enables faster processing of biomedical CT imagery as compared with prior methods described in the literature.