Proceedings. IEEE International Symposium on Biomedical Imaging最新文献

Spatially aligning two computed tomography (CT) scans of the lung using automated image registration techniques is a challenging task due to the deformable nature of the lung. However, existing deep-learning-based lung CT registration models are not trained with explicit anatomical knowledge. We propose the deformable anatomy-aware registration toolkit (DART), a masked autoencoder (MAE)-based approach, to improve the keypoint-supervised registration of lung CTs. Our method incorporates features from multiple decoders of networks trained to segment anatomical structures, including the lung, ribs, vertebrae, lobes, vessels, and airways, to ensure that the MAE learns relevant features corresponding to the anatomy of the lung. The pretrained weights of the transformer encoder and patch embeddings are then used as the initialization for the training of downstream registration. We compare DART to existing state-of-the-art registration models. Our experiments show that DART outperforms the baseline models (Voxelmorph, ViT-V-Net, and MAE-TransRNet) in terms of target registration error of both corrField-generated keypoints with 17%, 13%, and 9% relative improvement, respectively, and bounding box centers of nodules with 27%, 10%, and 4% relative improvement, respectively. Our implementation is available at https://github.com/yunzhengzhu/DART.

Multiplex immunofluorescence (MxIF) imaging is a critical tool in biomedical research, offering detailed insights into cell composition and spatial context. As an example, DAPI staining identifies cell nuclei, while CD20 staining helps segment cell membranes in MxIF. However, a persistent challenge in MxIF is saturation artifacts, which hinder single-cell level analysis in areas with over-saturated pixels. Traditional gamma correction methods for fixing saturation are limited, often incorrectly assuming uniform distribution of saturation, which is rarely the case in practice. This paper introduces a novel approach to correct saturation artifacts from a data-driven perspective. We introduce a two-stage, high-resolution hybrid generative adversarial network (HDmixGAN), which merges unpaired (CycleGAN) and paired (pix2pixHD) network architectures. This approach is designed to capitalize on the available small-scale paired data and the more extensive unpaired data from costly MxIF data. Specifically, we generate pseudo-paired data from large-scale unpaired over-saturated datasets with a CycleGAN, and train a Pix2pixGAN using both small-scale real and large-scale synthetic data derived from multiple DAPI staining rounds in MxIF. This method was validated against various baselines in a downstream nuclei detection task, improving the F1 score by 6% over the baseline. This is, to our knowledge, the first focused effort to address multi-round saturation in MxIF images, offering a specialized solution for enhancing cell analysis accuracy through improved image quality. The source code and implementation of the proposed method are available at https://github.com/MASILab/DAPIArtifactRemoval.git.

Traditional deep learning approaches for breast cancer classification has predominantly concentrated on single-view analysis. In clinical practice, however, radiologists concurrently examine all views within a mammography exam, leveraging the inherent correlations in these views to effectively detect tumors. Acknowledging the significance of multi-view analysis, some studies have introduced methods that independently process mammogram views, either through distinct convolutional branches or simple fusion strategies, inadvertently leading to a loss of crucial inter-view correlations. In this paper, we propose an innovative multi-view network exclusively based on transformers to address challenges in mammographic image classification. Our approach introduces a novel shifted window-based dynamic attention block, facilitating the effective integration of multi-view information and promoting the coherent transfer of this information between views at the spatial feature map level. Furthermore, we conduct a comprehensive comparative analysis of the performance and effectiveness of transformer-based models under diverse settings, employing the CBIS-DDSM and Vin-Dr Mammo datasets. Our code is publicly available at https://github.com/prithuls/MV-Swin-T.

The reconstruction of human visual inputs from brain activity, particularly through functional Magnetic Resonance Imaging (fMRI), holds promising avenues for unraveling the mechanisms of the human visual system. Despite the significant strides made by deep learning methods in improving the quality and interpretability of visual reconstruction, there remains a substantial demand for high-quality, long-duration, subject-specific 7-Tesla fMRI experiments. The challenge arises in integrating diverse smaller 3-Tesla datasets or accommodating new subjects with brief and low-quality fMRI scans. In response to these constraints, we propose a novel framework that generates enhanced 3T fMRI data through an unsupervised Generative Adversarial Network (GAN), leveraging unpaired training across two distinct fMRI datasets in 7T and 3T, respectively. This approach aims to overcome the limitations of the scarcity of high-quality 7-Tesla data and the challenges associated with brief and low-quality scans in 3-Tesla experiments. In this paper, we demonstrate the reconstruction capabilities of the enhanced 3T fMRI data, highlighting its proficiency in generating superior input visual images compared to data-intensive methods trained and tested on a single subject.

Physics-driven deep learning (PD-DL) methods have gained popularity for improved reconstruction of fast MRI scans. Though supervised learning has been used in early works, there has been a recent interest in unsupervised learning methods for training PD-DL. In this work, we take inspiration from statistical image processing and compressed sensing (CS), and propose a novel convex loss function as an alternative learning strategy. Our loss function evaluates the compressibility of the output image while ensuring data fidelity to assess the quality of reconstruction in versatile settings, including supervised, unsupervised, and zero-shot scenarios. In particular, we leverage the reweighted $l_1$ norm that has been shown to approximate the $l_0$ norm for quality evaluation. Results show that the PD-DL networks trained with the proposed loss formulation outperform conventional methods, while maintaining similar quality to PD-DL models trained using existing supervised and unsupervised techniques.

High-resolution full lung CT scans now enable the detailed segmentation of airway trees up to the 6th branching generation. The airway binary masks display very complex tree structures that may encode biological information relevant to disease risk and yet remain challenging to exploit via traditional methods such as meshing or skeletonization. Recent clinical studies suggest that some variations in shape patterns and caliber of the human airway tree are highly associated with adverse health outcomes, including all-cause mortality and incident COPD. However, quantitative characterization of variations observed on CT segmented airway tree remain incomplete, as does our understanding of the clinical and developmental implications of such. In this work, we present an unsupervised deep-learning pipeline for feature extraction and clustering of human airway trees, learned directly from projections of 3D airway segmentations. We identify four reproducible and clinically distinct airway sub-types in the MESA Lung CT cohort.

Resting-sate fMRI (rs-fMRI) has emerged as a viable tool to localize the epileptogenic zone (EZ) in medication refractory focal epilepsy patients. However, due to clinical protocol, datasets with reliable labels for the EZ are scarce. Some studies have used the entire resection area from post-operative structural T1 scans to act as the ground truth EZ labels during training and testing. These labels are subject to noise, as usually the resection area will be larger than the actual EZ tissue. We develop a mathematical framework for characterizing noisy labels in EZ localization. We use a multi-task deep learning framework to identify both the probability of a noisy label as well as the localization prediction for each ROI. We train our framework on a simulated dataset derived from the Human Connectome Project and evaluate it on both the simulated and a clinical epilepsy dataset. We show superior localization performance in our method against published localization networks on both the real and simulated dataset.

Robust quantification of pulmonary emphysema on computed tomography (CT) remains challenging for large-scale research studies that involve scans from different scanner types and for translation to clinical scans. Although the domain shifts in different CT scanners are subtle compared to shifts existing in other modalities (e.g., MRI) or cross-modality, emphysema is highly sensitive to it. Such subtle difference limits the application of general domain adaptation methods, such as image translation-based methods, as the contrast difference is too subtle to be distinguished. Existing studies have explored several directions to tackle this challenge, including density correction, noise filtering, regression, hidden Markov measure field (HMMF) model-based segmentation, and volume-adjusted lung density. Despite some promising results, previous studies either required a tedious workflow or eliminated opportunities for downstream emphysema subtyping, limiting efficient adaptation on a large-scale study. To alleviate this dilemma, we developed an end-to-end deep learning framework based on an existing HMMF segmentation framework. We first demonstrate that a regular UNet cannot replicate the existing HMMF results because of the lack of scanner priors. We then design a novel domain attention block, a simple yet efficient cross-modal block to fuse image visual features with quantitative scanner priors (a sequence), which significantly improves the results.

Physics-driven deep learning (PD-DL) has become a powerful tool for accelerated MRI. Recent developments have also developed unsupervised learning for PD-DL, including self-supervised learning. However, at very high acceleration rates, such approaches show performance deterioration. In this study, we propose to use cyclic-consistency (CC) to improve self-supervised learning for highly accelerated MRI. In our proposed CC, simulated measurements are obtained by undersampling the network output using patterns drawn from the same distribution as the true one. The reconstructions of these simulated measurements are obtained using the same network, which are then compared to the acquired data at the true sampling locations. This CC approach is used in conjunction with a masking-based self-supervised loss. Results show that the proposed method can substantially reduce aliasing artifacts at high acceleration rates, including rate 6 and 8 fastMRI knee imaging and 20-fold HCP-style fMRI.

Alzheimer's Disease (AD) is a neurodegenerative disorder characterized by progressive cognitive degeneration and motor impairment, affecting millions worldwide. Mapping the progression of AD is crucial for early detection of loss of brain function, timely intervention, and development of effective treatments. However, accurate measurements of disease progression are still challenging at present. This study presents a novel approach to understanding the heterogeneous pathways of AD through longitudinal biomarker data from medical imaging and other modalities. We propose an analytical pipeline adopting two popular machine learning methods from the single-cell transcriptomics domain, PHATE and Slingshot, to project multimodal biomarker trajectories to a low-dimensional space. These embeddings serve as our pseudotime estimates. We applied this pipeline to the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset to align longitudinal data across individuals at various disease stages. Our approach mirrors the technique used to cluster single-cell data into cell types based on developmental timelines. Our pseudotime estimates revealed distinct patterns of disease evolution and biomarker changes over time, providing a deeper understanding of the temporal dynamics of AD. The results show the potential of the approach in the clinical domain of neurodegenerative diseases, enabling more precise disease modeling and early diagnosis.