Learning deformable registration of medical images with anatomical constraints

Mansilla, Lucas; Milone, Diego H.; Ferrante, Enzo

doi:10.1016/j.neunet.2020.01.023

Cited by 68 publications

(57 citation statements)

References 46 publications

(79 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This illustrates the limits of simplistic metrics to reflect anatomic improvements in the image generation process. It is expected that more specific losses based on learnt representations could reflect this improvement [20]. The U-Net and the adversarial networks fail to generalize the upper section of the body where less training data were available.…”

Section: Discussionmentioning

confidence: 99%

Whole Body Positron Emission Tomography Attenuation Correction Map Synthesizing using 3D Deep Generative Adversarial Networks

Colmeiro¹,

Verrastro²,

Minsky³

et al. 2020

Preprint

View full text Add to dashboard Cite

Background: The correction of attenuation effects in Positron Emission Tomography (PET) imaging is fundamental to obtain a correct radiotracer distribution. However direct measurement of this attenuation map is not error-free and normally results in additional ionization radiation dose to the patient. Here, we propose to obtain the whole body attenuation map using a 3D U-Net generative adversarial network. The network is trained to learn the mapping from non attenuation corrected 18-F-fluorodeoxyglucose PET images to a synthetic Computerized Tomography (sCT) and also to label the input voxel tissue. The sCT image is further refined using an adversarial training scheme to recover higher frequency details and lost structures using context information. This work is trained and tested on public available datasets, containing several PET images from different scanners with different radiotracer administration and reconstruction modalities. The network is trained with 108 samples and validated on 10 samples.Results: The sCT generation was tested on 133 samples from 8 distinct datasets. The resulting mean absolute error of the network is 103 ± 18 HU and a peak signal to noise ratio of 18.6 ± 1.5 dB. The generated images show good correlation with the unknown structural information.Conclusions: The proposed deep learning topology is capable of generating whole body attenuation maps from uncorrected PET image data. Moreover, the method accuracy holds in the presence of data form multiple sources and modalities and is trained on publicly available datasets.

show abstract

Section: Discussionmentioning

confidence: 99%

Whole Body Positron Emission Tomography Attenuation Correction Map Synthesizing using 3D Deep Generative Adversarial Networks

Colmeiro¹,

Verrastro²,

Minsky³

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…The accuracy of the CTA-SPECT registration was evaluated for both of the registration methods with and without the whole heart constraint using the Dice similarity coefficient (30,31) to assess the overlap of the tracer and LV myocardium, and the Hausdorff distance (30,31) to assess the distance of the boundaries after performing the registration. The CTA-SPECT registration method was applied to both the minipig and patient data.…”

Section: Multimodality Image Registrationmentioning

confidence: 99%

Three-dimensional quantitative assessment of myocardial infarction via multimodality fusion imaging: methodology, validation, and preliminary clinical application

Tao

Liu

et al. 2021

Quant Imaging Med Surg

View full text Add to dashboard Cite

Background: The precise assessment of myocardial infarction (MI) is crucial both for therapeutic interventions in old MI and the development of new and effective techniques to repair injured myocardium.A novel method was developed to assess left ventricular (LV) quantitatively infarction through threedimensional (3D) multimodality fusion based on computed tomography angiography (CTA) and technetium-99m methoxyisobutylisonitrile ( 99m Tc-MIBI) single-photon emission computed tomography (SPECT) images. This study sought to develop a 3D quantitative method for MI for pre-clinical study and clinical application.Methods: Three months after the MI models were established in 20 minipigs, CTA and SPECT images were acquired separately, which were then aligned automatically with the constraints of the shape and the whole heart and LV myocardium position. Infarct ratios were quantified based on the 3D fusion images. The quantitative assessment was then experimentally validated via an ex vivo histology analysis using triphenyltetrazolium-chloride staining and subsequently applied to post-MI patients (n=8).Results: The location of an infarct identified by the SPECT was consistent with that identified by an ex vivo heart in a 3D space. Infarct size determined by CTA-SPECT was correlated with infarct size assessed by triphenyl-tetrazolium-chloride pathology {27.6% [interquartile range (IQR) 17.1-34.7%] vs. 24.1% (IQR 14.7-32.5%), r 2 =0.99, P<0.01}. In clinical cases, the CTA-SPECT 3D fusion quantitative results were significantly correlated with the quantitative perfusion SPECT results (r=0.976, P<0.01). Conclusions:The proposed 3D fusion quantitative assessment method provides reliable and intuitive evaluations of infarction. This novel quantification technique enables whole heart quantification for the pre-operation evaluation and post-diagnosis management of old MI patients. It could also be applied to the design of 3D-printed cardiac patches.

show abstract

“…As for x-ray images, there are six publicly available datasets, NLST [111], NIH ChestXray14 [116], OAI, JSRT [117], Montgomery County x-ray database [118] and Shenzhen Hospital x-ray database [118]. However, there are relatively few studies on x-ray image registration [25,80], compared with MRI and CT.…”

Section: Ultrasound Registration and X-ray Registrationmentioning

confidence: 99%

Deep learning in medical image registration

et al. 2020

View full text Add to dashboard Cite

Image registration is a fundamental task in multiple medical image analysis applications. With the advent of deep learning, there have been significant advances in algorithmic performance for various computer vision tasks in recent years, including medical image registration. The last couple of years have seen a dramatic increase in the development of deep learning-based medical image registration algorithms. Consequently, a comprehensive review of the current state-of-the-art algorithms in the field is timely, and necessary. This review is aimed at understanding the clinical applications and challenges that drove this innovation, analysing the functionality and limitations of existing approaches, and at providing insights to open challenges and as yet unmet clinical needs that could shape future research directions. To this end, the main contributions of this paper are: (a) discussion of all deep learning-based medical image registration papers published since 2013 with significant methodological and/or functional contributions to the field; (b) analysis of the development and evolution of deep learning-based image registration methods, summarising the current trends and challenges in the domain; and (c) overview of unmet clinical needs and potential directions for future research in deep learning-based medical image registration.

show abstract

Learning deformable registration of medical images with anatomical constraints

Cited by 68 publications

References 46 publications

Whole Body Positron Emission Tomography Attenuation Correction Map Synthesizing using 3D Deep Generative Adversarial Networks

Whole Body Positron Emission Tomography Attenuation Correction Map Synthesizing using 3D Deep Generative Adversarial Networks

Three-dimensional quantitative assessment of myocardial infarction via multimodality fusion imaging: methodology, validation, and preliminary clinical application

Deep learning in medical image registration

Contact Info

Product

Resources

About