Ch 9: Biomedical Imaging Informatics

• Biomedical imaging informatics is a growing discipline that recognizes the unique aspects of images as a core data type
• It shares similarities with natural language processing (NLP).
• The goal of imaging informatics methods is to extract information like anatomy and abnormalities for disease characterization.
• Imaging provides detailed information that complements electronic medical records data for capturing "electronic phenotypes".

Image Acquisition

• Involves delineating anatomic structure (anatomic/structural imaging) and determining tissue composition or function (functional imaging).
• Anatomic imaging aims to depict the structure of anatomic entities, while functional imaging focuses on determining tissue composition or function.
• Functional imaging shows the structure of the body and its function by observing changes in structure over time.
• Ultrasound and angiography are used to represent tissue composition, heart function, and blood flow.
• Molecular imaging depicts gene expression superimposed on structural images.
• Functional neuroimaging observes the electrical activity of neurons during cognitive tasks.
• Functional brain imaging modalities can be classified as image-based or non-image-based.
• Image-based functional data comes from scanners that generate relatively low-resolution volume arrays depicting spatially-localized activation.
• Electro-encephalography (EEG) or magnetoencephalography (MEG) are direct measures of electrical activity, using electromagnetic fields generated by neurons.
• The primary difference among imaging modalities is the type of energy source used to generate the images.
• Radiography (X-ray) is a projection technique, contrast is due to differences in tissue density and atomic number.
• Digital radiography (DR) directly creates digital radiographs, uses storage phosphor to replace X-rays.
• Computed Tomography (CT) uses X-rays to produce cross-sectional and volumetric images.
• Ultrasound uses high-frequency sound waves to image body structures, the basis of image generation is due to acoustical impedance.
• Magnetic Resonance Imaging (MRI) uses nuclear magnetic resonance (NMR) signals to create images from unpaired spinning charges in a magnetic field.
• Nuclear medicine imaging is a reverse of radiographic imaging, where the imaging source is inside the subject and projects out.
• Single-photon emission computed tomography (SPECT) produces a computed volumetric image that can be viewed and navigated in multiple planes.
• Positron Emission Tomography (PET) uses a special type of radioactive isotope that emits positrons, PET/CT integrates a PET scanner and CT.
• Image quality is a crucial aspect of medical imaging, with various parameters affecting the final image appearance, these include spatial resolution, contrast resolution, and temporal resolution.
• Contrast agents are introduced into the body to enhance imaging contrast of structures or fluids in medical imaging.
• Microscopic/cellular imaging is rapidly growing, with computational methods used to evaluate cell features.
• Pathology/tissue imaging has been revolutionized by the introduction of digital imaging and Picture Archiving and Communication Systems (PACS).
• Ophthalmologic imaging visualizes the retina for diagnosis and monitoring treatment response.
• Dermatologic imaging is becoming an important component of dermatology, with doctors taking photographs of patients with skin abnormalities.

Image Content Representation

• Consists of visual content and knowledge content.
• Visual content is the raw values of an image that can be directly accessed by computers.
• Knowledge content arises as the observer views the visual information in the image.
• Visual content is typically represented by a two-dimensional array of numbers (a bit map).
• Image metadata include the name of the patient, date the image was acquired, the slice thickness, and the modality used to acquire the image.
• Nomina Anatomica and Terminologia Anatomica provide a classification of officially sanctioned terms associated with macroscopic and microscopic anatomical structures.
• The Foundational Model of Anatomy (FMA) defines a comprehensive symbolic description of the structural organization of the body, consisting of over 75,000 concepts, represented by about 120,000 terms.
• RadLex is a controlled terminology for radiology for radiologists to communicate descriptions of imaging findings.
• The RadLex Playbook provides a standard system for naming radiology procedures based on atomic terms.
• The Logical Observation Identifiers, Names, and Codes (LOINC®) terminology includes radiology terms and has been created to represent the attributes of term names with an extensible set of values.
• There are two approaches to recording these assertions: narrative text (narrative text) and formal information model (formal information model).
• Semantic annotation methods are being developed to adapt the semantic content about images that would have been put into narrative text to be put in structured annotations compliant with the information model.
• The Annotation and Image Markup (AIM) schema captures information about image annotations and supports controlled terminologies, enabling semantic interoperability.
• The electronic Imaging Physician Annotation Device (ePAD) is a web-based image viewing and AIM-compliant annotation tool that allows users to draw image annotations in a manner they are accustomed to while viewing images.
• Digital atlases are spatial representations of anatomy, often combined with symbolic representations, created from 3-D image data from real subjects.
• An example of a 3-D brain atlas created from the Visible Human is Voxelman, where highly-detailed 3-D scenes are dynamically generated.
• The most widely used human brain atlas is the Talairach atlas, based on post-mortem sections from a 60-year-old woman.

Image Processing

• A form of signal processing that uses computational methods to produce an output image or a set of characteristics/parameters related to the image.
• Medical image processing uses tools similar to general image processing but presents unique challenges, including 3D body representation, multi-modalities, and fusion of information across several modalities.
• Image processing pipelines are generally built using one or more of the following fundamental image processing methods: global processing, image enhancement, image rendering/visualization, image quantitation, image segmentation, image registration, and image reasoning (e.g., classification).
• Global processing involves computations on the entire image without regard to specific regional content, aiming to enhance an image for human visualization or further analysis by the computer.
• Image enhancement uses global processing to improve the appearance of the image either for human use or for subsequent processing by computer.
• Image rendering and visualization are techniques used to create image displays, diagrams, or animations that provide a different perspective from raw images.
• Image quantitation is the process of extracting numerical parameters or deriving calculations from the image or from ROIs in the image.
• Quantitative image features are quantitative representations of visual signals contained in an image, such as texture, shapes, lesion margin characteristics, and image noise.
• Patch-based image representations and bag-of-features classification techniques have been proposed for processing image contents
• Radiomics is a field of computational methods that extract quantitative features from radiology images, focusing on identifying quantitative imaging indicators that predict important clinical outcomes.
• Deep learning is a rapidly emerging frontier in machine learning, providing a new approach to image analysis problems, particularly in medical imaging.
• Deep CNNs are particularly effective in object recognition and localization in natural images.
• Image segmentation involves automatically circumscribing regions within an image to generate ROIs, which usually correspond to anatomically meaningful structures or lesions.
• Popular segmentation techniques include region-based Vs. edge-based methods, knowledge-based Vs. data-driven methods, and combined methods.
• Model-based methods consider prior knowledge about the organ/medical images to be analyzed, while data-driven methods rely only on the specific analyzed image data.
• Clustering-based segmentation divides the image into a finite set of clusters/regions with similar statistics, smooth, and homogeneous in their content and representation.
• Tissue and lesion segmentation in brain MRI is a well-studied topic.

Image Interpretation and Computer Reasoning

• Image interpretation is a crucial stage in the clinical care process, where physicians provide a professional opinion on the presence of abnormalities in an image and their potential significance.
• Informatics methods can enhance radiological interpretation of images in two major ways: image retrieval systems and computer-based inference systems.
• Content-based image retrieval (CBIR) is the process of matching images using their visual content.
• CBIR methods are based on deriving quantitative characteristics from images and applying similarity metrics to search databases for similar images.
• Computer-based inference systems, specifically decision support systems, can help radiologists understand the biomedical import of information and provide guidance.
• There are two major approaches to computer-based inference using images: (1) using quantitative image features only (quantitative imaging computer inference systems) and (2) using knowledge associated with the images (knowledge-based computer inference systems).
• Quantitative imaging computer inference systems use quantifiable features extracted from medical images for decision support applications.
• There are three types of systems that make inferences using quantitative imaging data: computer-assisted detection (CAD), computer-assisted diagnosis (CADx), and computerized prediction systems.
• Computer-assisted detection (CAD) techniques for screening have been applied successfully to various types of images.
• CADx applications are used to identify suspicious regions in images and evaluate them for diagnosis or differential diagnosis.
• Computerized prediction using images aims to analyze disease characteristics in images and make predictions about the disease, such as life expectancy, treatment response, or future progression.
• Knowledge-based image inference systems, such as CAD and CADx, do not require processing radiological knowledge but use quantitative modeling of relationships between images features to diagnoses.
• Computer reasoning with ontologies can be performed through two methods: reasoning by ontology query and reasoning by logical inference.
• The Web Ontology Language (OWL) is recommended as a standard language for the Semantic Web and can capture knowledge by representing entities and their attributes.
• Knowledge-based reasoning with images is a method that uses ontologies as the knowledge source to process image content and derive inferences from them.
• Simulated knowledge-based reasoning methods could greatly enable functional evaluation of static abnormalities seen in medical imaging.
• Automated disease grading/staging uses image-based knowledge to automate grading of brain gliomas and staging of cancer based on imaging features detected by radiologists.
• Inferring the types of information users seek from images is another area where knowledge-based image reasoning systems can be developed.