Journal of Medical Imaging最新文献

Purpose: We provide a comparison of X-ray fluorescence emission tomography (XFET) and computed tomography (CT) for detecting low concentrations of gold nanoparticles (GNPs) in soft tissue and characterize the conditions under which XFET outperforms energy-integrating CT (EICT) and photon-counting CT (PCCT).

Approach: We compared dose-matched Monte Carlo XFET simulations and analytical fan-beam EICT and PCCT simulations. Each modality was used to image a numerical mouse phantom and contrast-depth phantom containing GNPs ranging from 0.05% to 4% by weight in soft tissue. Contrast-to-noise ratios (CNRs) of gold regions were compared among the three modalities, and XFET's detection limit was quantified based on the Rose criterion. A partial field-of-view (FOV) image was acquired for the phantom region containing 0.05% GNPs.

Results: For the mouse phantom, XFET produced superior CNR values ( $\mathrm{CNRs}=24.5$ , 21.6, and 3.4) compared with CT images obtained with both energy-integrating ( $\mathrm{CNR}=4.4$ , 4.6, and 1.5) and photon-counting ( $\mathrm{CNR}=6.5$ , 7.7, and 2.0) detection systems. More generally, XFET outperformed CT for superficial imaging depths ( $<28.75\mathrm{mm}$ ) for gold concentrations at and above 0.5%. XFET's surface detection limit was quantified as 0.44% for an average phantom dose of 16 mGy compatible with in vivo imaging. XFET's ability to image partial FOVs was demonstrated, and 0.05% gold was easily detected with an estimated dose of $\sim 81.6\mathrm{cGy}$ to a localized region of interest.

Conclusions: We demonstrate a proof of XFET's benefit for imaging low concentrations of gold at superficial depths and the feasibility of XFET for in vivo metal mapping in preclinical imaging tasks.

Purpose: Self-supervised pre-training can reduce the amount of labeled training data needed by pre-learning fundamental visual characteristics of the medical imaging data. We investigate several self-supervised training strategies for chest computed tomography exams and their effects on downstream applications.

Approach: We benchmark five well-known self-supervision strategies (masked image region prediction, next slice prediction, rotation prediction, flip prediction, and denoising) on 15 M chest computed tomography (CT) slices collected from four sites of the Mayo Clinic enterprise, United States. These models were evaluated for two downstream tasks on public datasets: pulmonary embolism (PE) detection (classification) and lung nodule segmentation. Image embeddings generated by these models were also evaluated for prediction of patient age, race, and gender to study inherent biases in models' understanding of chest CT exams.

Results: The use of pre-training weights especially masked region prediction-based weights, improved performance, and reduced computational effort needed for downstream tasks compared with task-specific state-of-the-art (SOTA) models. Performance improvement for PE detection was observed for training dataset sizes as large as $\sim 380K$ with a maximum gain of 5% over SOTA. The segmentation model initialized with pre-training weights learned twice as fast as the randomly initialized model. While gender and age predictors built using self-supervised training weights showed no performance improvement over randomly initialized predictors, the race predictor experienced a 10% performance boost when using self-supervised training weights.

Conclusion: We released self-supervised models and weights under an open-source academic license. These models can then be fine-tuned with limited task-specific annotated data for a variety of downstream imaging tasks, thus accelerating research in biomedical imaging informatics.

Purpose: Cells are building blocks for human physiology; consequently, understanding the way cells communicate, co-locate, and interrelate is essential to furthering our understanding of how the body functions in both health and disease. Hematoxylin and eosin (H&E) is the standard stain used in histological analysis of tissues in both clinical and research settings. Although H&E is ubiquitous and reveals tissue microanatomy, the classification and mapping of cell subtypes often require the use of specialized stains. The recent CoNIC Challenge focused on artificial intelligence classification of six types of cells on colon H&E but was unable to classify epithelial subtypes (progenitor, enteroendocrine, goblet), lymphocyte subtypes (B, helper T, cytotoxic T), and connective subtypes (fibroblasts). We propose to use inter-modality learning to label previously un-labelable cell types on H&E.

Approach: We took advantage of the cell classification information inherent in multiplexed immunofluorescence (MxIF) histology to create cell-level annotations for 14 subclasses. Then, we performed style transfer on the MxIF to synthesize realistic virtual H&E. We assessed the efficacy of a supervised learning scheme using the virtual H&E and 14 subclass labels. We evaluated our model on virtual H&E and real H&E.

Results: On virtual H&E, we were able to classify helper T cells and epithelial progenitors with positive predictive values of $0.34\pm 0.15$ (prevalence $0.03\pm 0.01$ ) and $0.47\pm 0.1$ (prevalence $0.07\pm 0.02$ ), respectively, when using ground truth centroid information. On real H&E, we needed to compute bounded metrics instead of direct metrics because our fine-grained virtual H&E predicted classes had to be matched to the closest available parent classes in the coarser labels from the real H&E dataset. For the real H&E, we could classify bounded metrics for the helper T cells and epithelial progenitors with upper bound positive predictive values of $0.43\pm 0.03$ (parent class prevalence 0.21) and $0.94\pm 0.02$ (parent class prevalence 0.49) when using ground truth centroid information.

Conclusions: This is the first work to provide cell type classification for helper T and epithelial progenitor nuclei on H&E.

Purpose: Deformable image registration establishes non-linear spatial correspondences between fixed and moving images. Deep learning-based deformable registration methods have been widely studied in recent years due to their speed advantage over traditional algorithms as well as their better accuracy. Most existing deep learning-based methods require neural networks to encode location information in their feature maps and predict displacement or deformation fields through convolutional or fully connected layers from these high-dimensional feature maps. We present vector field attention (VFA), a novel framework that enhances the efficiency of the existing network design by enabling direct retrieval of location correspondences.

Approach: VFA uses neural networks to extract multi-resolution feature maps from the fixed and moving images and then retrieves pixel-level correspondences based on feature similarity. The retrieval is achieved with a novel attention module without the need for learnable parameters. VFA is trained end-to-end in either a supervised or unsupervised manner.

Results: We evaluated VFA for intra- and inter-modality registration and unsupervised and semi-supervised registration using public datasets as well as the Learn2Reg challenge. VFA demonstrated comparable or superior registration accuracy compared with several state-of-the-art methods.

Conclusions: VFA offers a novel approach to deformable image registration by directly retrieving spatial correspondences from feature maps, leading to improved performance in registration tasks. It holds potential for broader applications.

Purpose: eXtended Reality (XR) technology, including virtual reality (VR), augmented reality (AR), and mixed reality (MR), is a growing field in healthcare. Each modality offers unique benefits and drawbacks for medical education, simulation, and clinical care. We review current studies to understand how XR technology uses medical imaging to enhance surgical diagnostics, planning, and performance. We also highlight current limitations and future directions.

Approach: We reviewed the literature on immersive XR technologies for surgical planning and intraoperative augmentation, excluding studies on telemedicine and 2D video-based training. We cited publications highlighting XR's advantages and limitations in these categories.

Results: A review of 556 papers on XR for medical imaging in surgery yielded 155 relevant papers reviewed utilizing the aid of chatGPT. XR technology may improve procedural times, reduce errors, and enhance surgical workflows. It aids in preoperative planning, surgical navigation, and real-time data integration, improving surgeon ergonomics and enabling remote collaboration. However, adoption faces challenges such as high costs, infrastructure needs, and regulatory hurdles. Despite these, XR shows significant potential in advancing surgical care.

Conclusions: Immersive technologies in healthcare enhance visualization and understanding of medical conditions, promising better patient outcomes and innovative treatments but face adoption challenges such as cost, technological constraints, and regulatory hurdles. Addressing these requires strategic collaborations and improvements in image quality, hardware, integration, and training.

Purpose: The critical time between stroke onset and treatment was targeted for reduction by integrating physiological imaging into the angiography suite, potentially improving clinical outcomes. The evaluation was conducted to compare C-Arm cone beam CT perfusion (CBCTP) with multi-detector CT perfusion (MDCTP) in patients with acute ischemic stroke (AIS).

Approach: Thirty-nine patients with anterior circulation AIS underwent both MDCTP and CBCTP. Imaging results were compared using an in-house algorithm for CBCTP map generation and RAPID for post-processing. Blinded neuroradiologists assessed images for quality, diagnostic utility, and treatment decision support, with non-inferiority analysis (two one-sided tests for equivalence) and inter-reviewer consistency (Cohen's kappa).

Results: The mean time from MDCTP to angiography suite arrival was $50\pm 34\min$ , and that from arrival to the first CBCTP image was $21\pm 8\min$ . Stroke diagnosis accuracies were 96% [93%, 97%] with MDCTP and 91% [90%, 93%] with CBCTP. Cohen's kappa between observers was 0.86 for MDCTP and 0.90 for CBCTP, showing excellent inter-reader consistency. CBCTP's scores for diagnostic utility, mismatch pattern detection, and treatment decisions were noninferior to MDCTP scores (alpha = 0.05) within 20% of the range. MDCTP was slightly superior for image quality and artifact score (1.8 versus 2.3, $p<0.01$ ).

Conclusions: In this small paper, CBCTP was noninferior to MDCTP, potentially saving nearly an hour per patient if they went directly to the angiography suite upon hospital arrival.

Purpose: The BRCA1-associated protein 1 (BAP1) gene is of great interest because somatic (BAP1) mutations are the most common alteration associated with pleural mesothelioma (PM). Further, germline mutation of the BAP1 gene has been linked to the development of PM. This study aimed to explore the potential of radiomics on computed tomography scans to identify somatic BAP1 gene mutations and assess the feasibility of radiomics in future research in identifying germline mutations.

Approach: A cohort of 149 patients with PM and known somatic BAP1 mutation status was collected, and a previously published deep learning model was used to first automatically segment the tumor, followed by radiologist modifications. Image preprocessing was performed, and texture features were extracted from the segmented tumor regions. The top features were selected and used to train 18 separate machine learning models using leave-one-out cross-validation (LOOCV). The performance of the models in distinguishing between BAP1-mutated (BAP1+) and BAP1 wild-type (BAP1-) tumors was evaluated using the receiver operating characteristic area under the curve (ROC AUC).

Results: A decision tree classifier achieved the highest overall AUC value of 0.69 (95% confidence interval: 0.60 and 0.77). The features selected most frequently through the LOOCV were all second-order (gray-level co-occurrence or gray-level size zone matrices) and were extracted from images with an applied transformation.

Conclusions: This proof-of-concept work demonstrated the potential of radiomics to differentiate among BAP1+/- in patients with PM. Future work will extend these methods to the assessment of germline BAP1 mutation status through image analysis for improved patient prognostication.

[This corrects the article DOI: 10.1117/1.JMI.11.6.062606.].

Purpose: The rapid development of highly multiplexed microscopy has enabled the study of cells embedded within their native tissue. The rich spatial data provided by these techniques have yielded exciting insights into the spatial features of human disease. However, computational methods for analyzing these high-content images are still emerging; there is a need for more robust and generalizable tools for evaluating the cellular constituents and stroma captured by high-plex imaging. To address this need, we have adapted spectral angle mapping-an algorithm developed for hyperspectral image analysis-to compress the channel dimension of high-plex immunofluorescence (IF) images.

Approach: Here, we present pseudo-spectral angle mapping (pSAM), a robust and flexible method for determining the most likely class of each pixel in a high-plex image. The class maps calculated through pSAM yield pixel classifications which can be combined with instance segmentation algorithms to classify individual cells.

Results: In a dataset of colon biopsies imaged with a 13-plex staining panel, 16 pSAM class maps were computed to generate pixel classifications. Instance segmentations of cells with Cellpose2.0 ( $F1$ -score of $0.83\pm 0.13$ ) were combined with these class maps to provide cell class predictions for 13 cell classes. In addition, in a separate unseen dataset of kidney biopsies imaged with a 44-plex staining panel, pSAM plus Cellpose2.0 ( $F1$ -score of $0.86\pm 0.11$ ) detected a diverse set of 38 classes of structural and immune cells.

Conclusions: In summary, pSAM is a powerful and generalizable tool for evaluating high-plex IF image data and classifying cells in these high-dimensional images.