Machine Learning Approaches Towards Medical Images

Machine Learning Approaches Towards Medical Images

Gayathri S. P., Siva Shankar Ramasamy, Vijayalakshmi S.
DOI: 10.4018/978-1-6684-6523-3.ch006
(Individual Chapters)
No Current Special Offers


Clinical imaging relies heavily on the current medical services' framework to perform painless demonstrative therapy. It entails creating usable and instructive models of the human body's internal organs and structural systems for use in clinical evaluation. Its various varieties include signal-based techniques such as conventional X-ray, computed tomography (CT), magnetic resonance imaging (MRI), ultrasound (US) imaging, and mammography. Despite these clinical imaging techniques, clinical images are increasingly employed to identify various problems, particularly those that are upsetting the skin. Imaging and processing are the two distinct patterns of clinical imaging. To diagnose diseases, automatic segmentation using deep learning techniques in the field of clinical imaging is becoming vital for identifying evidence and measuring examples in clinical images. The fundamentals of deep learning techniques are discussed in this chapter along with an overview of successful implementations.
Chapter Preview


The creation of biological image processing algorithms has significantly advanced as a result. This has made it possible to create automated algorithms for information extraction through image analysis or evaluation. Segmentation, which splits the image into visually distinct parts with semantic meaning for the given problem, is the fundamental stage in this automated analysis. Each area often has consistent features regarding its color, texture, or grey level. Precise segmentation and discernible sections are crucial for additional research that may require determining the homogeneity levels of texture or layer thickness. There may occasionally be several items of the same class in the image. Instance segmentation is the separation of regions containing items belonging to the same class while disregarding other courses, as opposed to semantic segmentation, in which objects belonging to the same type are not separated, but various categories are. Three categories can be used to categorize all image segmentation approaches manual segmentation (MS), semi-automatic segmentation, and fully automatic segmentation techniques. For MS methods to properly annotate each picture pixel, subject matter experts must first identify the region of interest (ROI) and then draw exact boundaries around the ROI. MS is essential for the advancement of semi-automatic and utterly automatic segmentation algorithms since it gives the tagged ground truth pictures. MS takes a lot of time and is only practical for tiny image datasets. When it comes to high-quality images, the high resolution could mean that the edges are no longer clearly defined. As a result, even little changes in the ROI boundary's pixel selection can cause significant errors. Another problem with manual segmentation is that it is subjective because it depends on the knowledge and experience of the expert, and as a result, there is frequently a lot of variation among and within experts.

Complete Chapter List

Search this Book: