Automated Quality Enhancement, Modelling and Management of Diagnostic Scan Images With Ai Techniques

Abstract

Diagnostic scanning is extensively used for investigation of internal organ-related ailments and managing patient care. With the proliferation of imaging-based di- agnostic procedures in healthcare, patient-specific scan images constitute huge volumes of data, thus creating a need for automated healthcare information man- agement systems (HIMS) to facilitate their efficient organization and manage- ment, and for supporting clinical decision support applications. Medical images often require varied processing for enabling effective representation and modeling for building higher-level decision-support applications. One of the critical gaps in automated systems is limited attention to certain standards in meeting the quality of the scanned images. Compounding this problem is the availability of multi-vendor, non-standard scan resolution machines and also ill-trained medi- cal technicians. Automatically making computers understand the content of an image and offering a reasonable description in natural language has gained more attention recently in computer vision and natural language processing research communities. The caption prediction task in the medical domain is thus very relevant, as it aims to generate textual descriptions of the images, which can be used to improve indexing mechanisms in HIMS. The focus of the research work presented in this thesis is on building an ef- fective framework for medical image representation, modeling and management, for enabling advanced clinical applications like similarity based diagnostics, deci- sion support, etc. In clinical diagnosis, diagnostic images that are obtained from the scanning devices serve as preliminary evidence for further investigation in the process of delivering quality healthcare. However, often the medical image may contain fault artifacts introduced due to noise, blur and faulty equipment. The reason for this may be low-quality or older scanning devices, the test environment or technician’s lack of training etc, however, the net result is that the process of fast and accurate diagnosis is hampered. Towards this, automated image quality improvement approaches are adapted and benchmarked for the task of medical image quality enhancement through super-resolution. iii iv To design approaches for leveraging the enhanced medical images for further analysis and modeling for supporting applications like categorization, retrieval and automated captioning using machine learning and deep learning techniques, the concept of Content-based Medical Image Retrieval (CBMIR) systems is in- corporated. The CBMIR system designed can model heterogeneous views, body orientation, etc for supporting similar image retrieval for diagnosis. In diagnos- tic medical images, the patient body orientation or view of the scanning posture like anterior or frontal view, posterior or back view and the lateral or side views, also known as left lateral or right lateral can be used during scanning. However, computer-aided diagnosis systems often do not provide this piece of header in- formation of the image. Hence, image orientation identification is essential for qualitative and quantitative analysis in diagnostic applications. If such patient body orientations are not recorded or are documented using an incorrect label, automated system indexing may be inconsistent, and may also result in improper interpretation by computers and radiologists. Thus, a learnable neural model for accurately identifying the view positions of different organs of the body is proposed and designed. For a radiologist to delineate the imaging study’s findings/observations as a textual report is a manual, time consuming and tedious task, further exacerbated by the volume of generated images. Automated methods for radiographic image examination for identifying abnormalities and generating reliable radiology report are thus a critical requirement in clinical workflow management applications. The features extracted using neural network architectures are used to automatically generate the diagnosis medical report for scanned images, thus providing a way to build a robust medical imaging application for quality diagnosis. The promising achieved results underscore the performance of the approaches designed in this research and reveal much scope for adaptation in the healthcare field for improving the quality of healthcare delivery and management.