High-resolution (HR) 3D fetal brain magnetic resonance imaging (MRI) volume reconstruction from multiple motion-corrupted stacks of 2D thick slices is crucial for clinical diagnosis and quantitative analysis. Reliable sliceto-volume registration (SVR)-based motion correction and super-resolution reconstruction (SRR) methods are essential for high-quality isotropic volume reconstruction. Deep learning (DL) has demonstrated potential in enhancing motion correction and SRR when compared to conventional methods. However, current supervised DL methods for SVR and SRR require external large-scale training datasets, which are difficult to obtain in clinical fetal MRI settings. To address these issues, we propose an unsupervised iterative SVR-SRR framework for isotropic HR volume reconstruction without using external databases. Specifically, we formulate the SVR process as a function that maps a thick 2D input slice and a target 3D volume to a rigid transformation matrix, which aligns the slice to the underlying location in the target volume. The function is parameterized by a convolutional neural network, which is trained by minimizing the difference between the volume slicing at the predicted position and the input slice. For the SRR process, we utilize a decoding network possessing a deep image prior framework with a comprehensive image degradation model to generate the HR volume. The decoding network, utilizing a forward degradation model, offers a local consistency prior to guide the reconstruction of HR volumes from input slices of individual subjects. Comprehensive experiments conducted on large-magnitude motion-corrupted simulation data and clinical data demonstrate the superior performance of the proposed framework over state-of-theart fetal brain reconstruction frameworks.
MedIA 2024
×
COLLATOR: Consistent spatial-temporal longitudinal atlas construction via implicit neural representation
Longitudinal brain atlases that present brain development trend along time, are essential tools for brain development studies. However, conventional methods construct these atlases by independently averaging brain images from different individuals at discrete time points. This approach could introduce temporal inconsistencies due to variations in ontogenetic trends among samples, potentially affecting accuracy of brain developmental characteristic analysis. In this paper, we propose an implicit neural representation (INR)-based framework to improve the temporal consistency in longitudinal atlases. We treat temporal inconsistency as a 4-dimensional (4D) image denoising task, where the data consists of 3D spatial information and 1D temporal progression. We formulate the longitudinal atlas as an implicit function of the spatial-temporal coordinates, allowing structural inconsistency over the time to be considered as 3D image noise along age. Inspired by recent self-supervised denoising methods (e.g. Noise2Noise), our approach learns the noise-free and temporally continuous implicit function from inconsistent longitudinal atlas data. Finally, the time-consistent longitudinal brain atlas can be reconstructed by evaluating the denoised 4D INR function at critical brain developing time points. We evaluate our approach on three longitudinal brain atlases of different MRI modalities, demonstrating that our method significantly improves temporal consistency while accurately preserving brain structures. Additionally, the continuous functions generated by our method enable the creation of 4D atlases with higher spatial and temporal resolution.
NeurIPS 2023
×
Unsupervised Polychromatic Neural Representation for CT Metal Artifact Reduction
Qing Wu
, Lixuan Chen
, Ce Wang
, Hongjiang Wei
, S. Kevin Zhou
, Jingyi Yu
, Yuyao Zhang
Advances in Neural Information Processing Systems 2023
Emerging neural reconstruction techniques based on tomography (e.g., NeRF, NeAT, and NeRP) have started showing unique capabilities in medical imaging. In this work, we present a novel Polychromatic neural representation (Polyner) to tackle the challenging problem of CT imaging when metallic implants exist within the human body. The artifacts arise from the drastic variation of metal’s attenuation coefficients at various energy levels of the X-ray spectrum, leading to a nonlinear metal effect in CT measurements. Reconstructing CT images from metal-affected measurements hence poses a complicated nonlinear inverse problem where empirical models adopted in previous metal artifact reduction (MAR) approaches lead to signal loss and strongly aliased reconstructions. Polyner instead models the MAR problem from a nonlinear inverse problem perspective. Specifically, we first derive a polychromatic forward model to accurately simulate the nonlinear CT acquisition process. Then, we incorporate our forward model into the implicit neural representation to accomplish reconstruction. Lastly, we adopt a regularizer to preserve the physical properties of the CT images across different energy levels while effectively constraining the solution space. Our Polyner is an unsupervised method and does not require any external training data. Experimenting with multiple datasets shows that our Polyner achieves comparable or better performance than supervised methods on in-domain datasets while demonstrating significant performance improvements on out-of-domain datasets. To the best of our knowledge, our Polyner is the first unsupervised MAR method that outperforms its supervised counterparts.
IEEE ISBI 2023
×
ASSURED: A Self-supervised Deep Decoder Network for Fetus Brain MRI Reconstruction
Jiangjie Wu
, Lixuan Chen
, Zhenghao Li
, Rongpin Wang
, Hongjiang Wei
, Yuyao Zhang
20th IEEE International Symposium on Biomedical Imaging
High-resolution Magnetic Resonance Imaging (MRI) volume reconstruction from multiple arbitrary orientation motion-corrupted 2D slices is crucial for fetal brain MRI studies. Currently, most existing methods follow two-step approaches that iteratively perform slice to volume registration (SVR) and super-resolution reconstruction (SRR). However, the 3D volume reconstruction is often corrupted due to slice misalignment and brain anatomy blurring caused by severe motion during MR data collection, making the quantification challenging. To tackle these issues, we propose a novel learning-based self-supervised volume reconstruction technique that is robust to slice misalignment and motion artifacts. Specially, we combine a comprehensive forward model to present the complex image degradation process and an under-parameterized deep decoder structure to reduce the network overfitting with image artifacts caused by slice misalignment and motion. This methodology requires only one coarse SVR step in the whole reconstruction process and does not need any training dataset in SRR. We evaluated the performance of our technique on simulated MRI from brain atlas and on real clinical scanning fetus MR data. Experimental results demonstrated that the proposed approach achieved superior fetus brain reconstruction results compared with state-of-the-art methods
PIPPI@ MICCAI 2022
Oral
×
Continuous longitudinal fetus brain atlas construction via implicit neural representation
Longitudinal fetal brain atlas is a powerful tool for understanding and characterizing the complex process of fetus brain development. Existing fetus brain atlases are typically constructed by averaged brain images on discrete time points independently over time. Due to the differences in onto-genetic trends among samples at different time points, the resulting atlases suffer from temporal inconsistency, which may lead to estimating error of the brain developmental characteristic parameters along the timeline. To this end, we proposed a multi-stage deep-learning framework to tackle the time inconsistency issue as a 4D (3D brain volume + 1D age) image data denoising task. Using implicit neural representation, we construct a continuous and noise-free longitudinal fetus brain atlas as a function of the 4D spatial-temporal coordinate. Experimental results on two public fetal brain atlases (CRL and FBA-Chinese atlases) show that the proposed method can significantly improve the atlas temporal consistency while maintaining good fetus brain structure representation. In addition, the continuous longitudinal fetus brain atlases can also be extensively applied to generate finer 4D atlases in both spatial and temporal resolution.
