Search

Qualitative visual assessment of MRI scans is a key mechanism by which inflammation is assessed in clinical practice. For example, in axial spondyloarthritis (axSpA), visual assessment focuses on the identification of regions with increased signal in the bone marrow, known as bone marrow oedema (BMO), on water-sensitive images. The identification of BMO has an important role in the diagnosis, quantification and monitoring of disease in axSpA. However, BMO evaluation depends heavily on the experience and expertise of the image reader, creating substantial imprecision. Deep learning-based segmentation is a natural approach to addressing this imprecision, but purely automated solutions require large training sets that are not currently available, and deep learning solutions with limited data may not be sufficiently trustworthy for use in clinical practice. To address this, we propose a workflow for inflammation segmentation incorporating both deep learning and human input. With this 'human-machine cooperation' workflow, a preliminary segmentation is generated automatically by deep learning; a human reader then 'cleans' the segmentation by removing extraneous segmented voxels. The final cleaned segmentation defines the volume of hyperintense inflammation (VHI), which is proposed as a quantitative imaging biomarker (QIB) of inflammation load in axSpA. We implemented and evaluated the proposed human-machine workflow in a cohort of 29 patients with axSpA who had undergone prospective MRI scans before and after starting biologic therapy. The performance of the workflow was compared against purely visual assessment in terms of inter-observer/inter-method segmentation overlap, inter-observer agreement and assessment of response to biologic therapy. The human-machine workflow showed superior inter-observer segmentation overlap than purely manual segmentation (Dice score 0.84 versus 0.56). VHI measurements produced by the workflow showed similar or better inter-observer agreement than visual scoring, with similar response assessments. We conclude that the proposed human-machine workflow offers a mechanism to improve the consistency of inflammation assessment, and that VHI could be a valuable QIB of inflammation load in axSpA, as well as offering an exemplar of human-machine cooperation more broadly.

This paper proposes a task-driven computational framework for assessing diffusion MRI experimental designs which, rather than relying on parameter-estimation metrics, directly measures quantitative task performance. Traditional computational experimental design (CED) methods may be ill-suited to experimental tasks, such as clinical classification, where outcome does not depend on parameter-estimation accuracy or precision alone. Current assessment metrics evaluate experiments' ability to faithfully recover microstructural parameters rather than their task performance. The method we propose addresses this shortcoming. For a given MRI experimental design (protocol, parameter-estimation method, model, etc.), experiments are simulated start-to-finish and task performance is computed from receiver operating characteristic (ROC) curves and associated summary metrics (e.g. area under the curve (AUC)). Two experiments were performed: first, a validation of the pipeline's task performance predictions against clinical results, comparing in-silico predictions to real-world ROC/AUC; and second, a demonstration of the pipeline's advantages over traditional CED approaches, using two simulated clinical classification tasks. Comparison with clinical datasets validates our method's predictions of (a) the qualitative form of ROC curves, (b) the relative task performance of different experimental designs, and (c) the absolute performance (AUC) of each experimental design. Furthermore, we show that our method outperforms traditional task-agnostic assessment methods, enabling improved, more useful experimental design. Our pipeline produces accurate, quantitative predictions of real-world task performance. Compared to current approaches, such task-driven assessment is more likely to identify experimental designs that perform well in practice. Our method is not limited to diffusion MRI; the pipeline generalises to any task-based quantitative MRI application, and provides the foundation for developing future task-driven end-to end CED frameworks.

OBJECTIVE:To identify the whole-body MRI (WB-MRI) image type(s) with the highest value for assessment of multiple myeloma, in order to optimise acquisition protocols and read times. METHODS:Thirty patients with clinically-suspected MM underwent WB-MRI at 3 Tesla. Unenhanced Dixon images [fat-only (FO) and water-only (WO)], post contrast Dixon [fat-only plus contrast (FOC) and water-only plus contrast (WOC)] and diffusion weighted images (DWI) of the pelvis from all 30 patients were randomised and read by three experienced readers. For each image type, each reader identified and labelled all visible myeloma lesions. Each identified lesion was compared with a composite reference standard achieved by review of a complete imaging dataset by a further experienced consultant radiologist to determine truly positive lesions. Lesion count, true positives, sensitivity, and positive predictive value were determined. Time to read each scan set was recorded. Confidence for a diagnosis of myeloma was scored using a Likert scale. Conspicuity of focal lesions was assessed in terms of percent contrast and contrast to noise ratio (CNR). RESULTS:Lesion count, true positives, sensitivity and confidence scores were significantly higher when compared to other image types for DWI (P

To determine which of four Dixon image types [in-phase (IP), out-of-phase (OP), fat only (FO) and water-only (WO)] is most sensitive for detecting multiple myeloma (MM) focal lesions on whole body MRI (WB-MRI) images.Thirty patients with clinically-suspected MM underwent WB-MRI at 3 Tesla. Unenhanced IP, OP, FO and WO Dixon images were generated and read by four radiologists. On each image type, each radiologist identified and labelled all visible myeloma lesions in the bony pelvis. Each identified lesion was compared with a reference standard consisting of pre- and post-contrast Dixon and diffusion weighted imaging (read by a further consultant radiologist) to determine whether the lesion was truly positive. Lesion count, true positives, sensitivity, and positive predictive value were compared across the four Dixon image types.Lesion count, true positives, sensitivity and confidence scores were all significantly higher on FO images than on IP images (p>0.05).FO images are more sensitive than other Dixon image types for MM focal lesions, and should be preferentially read by radiologists to improve diagnostic accuracy and reporting efficiency.