Accelerated MRI using intelligent protocolling and subject-specific denoising applied to Alzheimer’s disease imaging

Abstract

Magnetic Resonance Imaging (MRI) is expensive and time-consuming. Protocol optimization to accelerate MRI requires local expertise since each MR sequence involves multiple configurable parameters that need optimization for contrast, acquisition time, and signal-to-noise ratio (SNR). The availability and access to technical training are limited in under-served regions, resulting in a scarcity of local expertise required to operate the hardware and perform MR examinations. Along with other cultural and temporal constraints, these factors contribute to the highly inefficient utilization of MRI services diminishing their clinical value. In this work, we extend our previous effort and demonstrate accelerated MRI via intelligent protocolling of the modified brain screen protocol, referred to as the Gold Standard (GS) protocol. We leverage deep learning-based contrast-specific image-denoising to improve the image quality of data acquired using the accelerated protocol. Since the SNR of MR acquisitions depends on the volume of the object being imaged, we demonstrate subject-specific (SS) image-denoising. Utilizing the accelerated protocol resulted in a 1.94x gain in imaging throughput over the GS protocol. The minimum /maximum PSNR gains (measured in dB) were 1.18/11.68 and 1.04/13.15, from the baseline and SS image-denoising models, respectively. Alzheimers Disease (AD) accounts for up to 60-80% of dementia cases and a global trend of longer lifespans has resulted in an increase in the prevalence of dementia/AD. Therefore, an accurate differential diagnosis of AD is crucial to determine the right course of treatment. The GS protocol constitutes 44.44% of the comprehensive AD imaging protocol defined by the European Prevention of Alzheimers Disease project. Therefore, we also demonstrate the potential for AD-imaging via automated volumetry of relevant brain anatomies whose atrophies have been shown to be reliable indicators of the onset of the disease. The volumetric measurements of the hippocampus and amygdala from the GS and accelerated protocols were in excellent agreement, as measured by the intra-class correlation coefficient. In conclusion, accelerated brain imaging with the potential for AD imaging was demonstrated, and image quality was recovered post-acquisition using DL-based image denoising models.

Developing and deploying deep learning models in brain magnetic resonance imaging: A review

“…4 Second is the acceleration of existing vendor-defined protocols, potentially relying on post acquisition methods to recover SNR. 83,84 This approach is inherently limited to a particular protocol and vendor. The emergence of physics-informed DL methods will allow researchers to develop models that are privy to the underlying physical phenomena, potentially resulting in improved interpretability because the outputs can be evaluated using existing task-specific knowledge.…”

“…98 Image denoising models improve SNR postacquisition. 83,84 Two common approaches to achieve image denoising are to either directly synthesize the denoised image, or to synthesize the residual from which the final denoised image can be obtained. In the first approach, the models are trained on pairs of noisy/clean images to optimize image quality, while avoiding blurring artifacts and retaining the anatomical structures present in the original image.…”

“…In this approach, once a vendor hardware of interest has been identified, imposing appropriate constraints on the cost function (e.g., slew rate) will facilitate easy implementation of the optimized pulse sequence on the chosen hardware 4 . Second is the acceleration of existing vendor‐defined protocols, potentially relying on post acquisition methods to recover SNR 83,84 . This approach is inherently limited to a particular protocol and vendor.…”

“…One workaround is to leverage a self‐supervised learning framework to synthesize LR images from HR data, thereby mitigating the requirement for image registration 98 . Image denoising models improve SNR postacquisition 83,84 . Two common approaches to achieve image denoising are to either directly synthesize the denoised image, or to synthesize the residual from which the final denoised image can be obtained.…”

Accelerated MRI using intelligent protocolling and subject-specific denoising applied to Alzheimer's disease imaging

“…MR Value is an initiative by the International Society of Magnetic Resonance in Medicine to measure the utility of MR Imaging in the context of constantly evolving healthcare economics ( https://www.ismrm.org/the-mr-value-initiative-phase-1/ ). We based our prior work on this premise and demonstrated preliminary results from MR Value-driven Autonomous MR Imaging, dubbed AMRI (Ravi and Geethanath, 2020 ; Ravi et al, 2020 ).…”