A Task-Based Approach to Adaptive and Multimodality Imaging

Abstract

Multimodality imaging is becoming increasingly important in medical imaging. Since the motivation for combining multiple imaging modalities is generally to improve diagnostic or prognostic accuracy, the benefits of multimodality imaging cannot be assessed through the display of example images. Instead, we must use objective, task-based measures of image quality to draw valid conclusions about system performance. In this paper, we will present a general framework for utilizing objective, taskbased measures of image quality in assessing multimodality and adaptive imaging systems. We introduce a classification scheme for multimodality and adaptive imaging systems and provide a mathematical description of the imaging chain along with block diagrams to provide a visual illustration. We show that the task-based methodology developed for evaluating single-modality imaging can be applied, with minor modifications, to multimodality and adaptive imaging. We discuss strategies for practical implementing of task-based methods to assess and optimize multimodality imaging systems.

Proximal ADMM for Multi-Channel Image Reconstruction in Spectral X-ray CT

“…Signal detection and estimation tasks are both highly relevant and important for evaluating image quality in K-edge imaging. Recent advancements in image quality assessment methods for multi-modality imaging [49] may likely be useful for this task.…”

Adaptive small-animal SPECT/CT

“…We are now completing a helical-scan cone-beam X-ray CT with adaptive features that will physically mate to the FastSPECT II system. Data from the CT will not only yield complementary anatomical information that provides context for SPECT images, but will also be used for multi-modality adaptive feedback rules as discussed below and in more detail in a topical issue of the Proceedings of the IEEE [2]. This CT system uses an X-ray source with a 6-μm focal spot, programmable tube voltage (≤ 130 kVp), and programmable tube current (≤ .5 mA), and a 2048×1000-pixel CMOS detector with 48-μm pitch.…”

“…We have derived expressions for the ideal linear observers for estimation and classification tasks in adaptive systems, and have suggested a wide variety of possible adaptation rules [1,2]. In essence, the image from the first modality is used to narrow down the ensemble of subjects from which the particular subject could have been drawn, and the second modality is then optimized for this new ensemble, which we refer to as the posterior ensemble because it is constructed after acquiring the data from the first modality.…”

“…In recent work, we have investigated the concept of adapting biomedical-imager configuration to optimize task-based image quality for a particular imaging subject rather than a population of subjects [1][2][3]. This involves acquiring and analyzing preliminary scout data and making automated decisions for setting the adjustable parameters in the system, which for SPECT may include choice of apertures, selection of magnification and field of view as determined by radial camera and aperture locations, and determination of overall acquisition trajectory [1].…”

“…In addition, we have developed a theoretical framework for the computations required to adapt a system optimally to a particular subject and task, and we have analyzed the benefits of creating feedback rules that involve data from a complementary modality, such as X-ray computed tomography (CT) [2]. These preliminary studies have led us to the conclusion that adaptive SPECT/CT represents a significant advance in molecular imaging that will have a positive impact in both laboratory and clinical settings.…”