Sinogram domain angular upsampling of sparse-view micro-CT with dense residual hierarchical transformer and attention-weighted loss.

Reduced angular sampling is a key strategy for increasing scanning efficiency of micron-scale computed tomography (micro-CT). Despite boosting throughput, this strategy introduces noise and extrapolation artifacts due to undersampling. In this work, we present a solution to this issue, by proposing a novel Dense Residual Hierarchical Transformer (DRHT) network to recover high-quality sinograms from 2×, 4× and 8× undersampled scans. DRHT is trained to utilize limited information available from sparsely angular sampled scans and once trained, it can be applied to recover higher-resolution sinograms from shorter scan sessions. Our proposed DRHT model aggregates the benefits of a hierarchical- multi-scale structure along with the combination of local and global feature extraction through dense residual convolutional blocks and non-overlapping window transformer blocks respectively. We also propose a novel noise-aware loss function named KL-L1 to improve sinogram restoration to full resolution. KL-L1, a weighted combination of pixel-level and distribution-level cost functions, leverages inconsistencies in noise distribution and uses learnable spatial weight maps to improve the training of the DRHT model. We present ablation studies and evaluations of our method against other state-of-the-art (SOTA) models over multiple datasets. Our proposed DRHT network achieves an average increase in peak signal to noise ratio (PSNR) of 17.73 dB and a structural similarity index (SSIM) of 0.161, for 8× upsampling, across the three diverse datasets, compared to their respective Bicubic interpolated versions. This novel approach can be utilized to decrease radiation exposure to patients and reduce imaging time for large-scale CT imaging projects.

AI Chatbots in Clinical Laboratory Medicine: Foundations and Trends.

BACKGROUND: Artificial intelligence (AI) conversational agents, or chatbots, are computer programs designed to simulate human conversations using natural language processing. They offer diverse functions and applications across an expanding range of healthcare domains. However, their roles in laboratory medicine remain unclear, as their accuracy, repeatability, and ability to interpret complex laboratory data have yet to be rigorously evaluated. CONTENT: This review provides an overview of the history of chatbots, two major chatbot development approaches, and their respective advantages and limitations. We discuss the capabilities and potential applications of chatbots in healthcare, focusing on the laboratory medicine field. Recent evaluations of chatbot performance are presented, with a special emphasis on large language models such as the Chat Generative Pre-trained Transformer in response to laboratory medicine questions across different categories, such as medical knowledge, laboratory operations, regulations, and interpretation of laboratory results as related to clinical context. We analyze the causes of chatbots' limitations and suggest research directions for developing more accurate, reliable, and manageable chatbots for applications in laboratory medicine. SUMMARY: Chatbots, which are rapidly evolving AI applications, hold tremendous potential to improve medical education, provide timely responses to clinical inquiries concerning laboratory tests, assist in interpreting laboratory results, and facilitate communication among patients, physicians, and laboratorians. Nevertheless, users should be vigilant of existing chatbots' limitations, such as misinformation, inconsistencies, and lack of human-like reasoning abilities. To be effectively used in laboratory medicine, chatbots must undergo extensive training on rigorously validated medical knowledge and be thoroughly evaluated against standard clinical practice.

Quantitative Geometric Modeling of Blood Cells from X-ray Histotomograms of Whole Zebrafish Larvae.

Tissue phenotyping is foundational to understanding and assessing the cellular aspects of disease in organismal context and an important adjunct to molecular studies in the dissection of gene function, chemical effects, and disease. As a first step toward computational tissue phenotyping, we explore the potential of cellular phenotyping from 3-Dimensional (3D), 0.74 µm isotropic voxel resolution, whole zebrafish larval images derived from X-ray histotomography, a form of micro-CT customized for histopathology. As proof of principle towards computational tissue phenotyping of cells, we created a semi-automated mechanism for the segmentation of blood cells in the vascular spaces of zebrafish larvae, followed by modeling and extraction of quantitative geometric parameters. Manually segmented cells were used to train a random forest classifier for blood cells, enabling the use of a generalized cellular segmentation algorithm for the accurate segmentation of blood cells. These models were used to create an automated data segmentation and analysis pipeline to guide the steps in a 3D workflow including blood cell region prediction, cell boundary extraction, and statistical characterization of 3D geometric and cytological features. We were able to distinguish blood cells at two stages in development (4- and 5-days-post-fertilization) and wild-type vs. polA2 huli hutu ( hht ) mutants. The application of geometric modeling across cell types to and across organisms and sample types may comprise a valuable foundation for computational phenotyping that is more open, informative, rapid, objective, and reproducible.

BBox-Guided Segmentor: Leveraging expert knowledge for accurate stroke lesion segmentation using weakly supervised bounding box prior.

Stroke is one of the leading causes of death and disability in the world. Despite intensive research on automatic stroke lesion segmentation from non-invasive imaging modalities including diffusion-weighted imaging (DWI), challenges remain such as a lack of sufficient labeled data for training deep learning models and failure in detecting small lesions. In this paper, we propose BBox-Guided Segmentor, a method that significantly improves the accuracy of stroke lesion segmentation by leveraging expert knowledge. Specifically, our model uses a very coarse bounding box label provided by the expert and then performs accurate segmentation automatically. The small overhead of having the expert provide a rough bounding box leads to large performance improvement in segmentation, which is paramount to accurate stroke diagnosis. To train our model, we employ a weakly-supervised approach that uses a large number of weakly-labeled images with only bounding boxes and a small number of fully labeled images. The scarce fully labeled images are used to train a generator segmentation network, while adversarial training is used to leverage the large number of weakly-labeled images to provide additional learning signals. We evaluate our method extensively using a unique clinical dataset of 99 fully labeled cases (i.e., with full segmentation map labels) and 831 weakly labeled cases (i.e., with only bounding box labels), and the results demonstrate the superior performance of our approach over state-of-the-art stroke lesion segmentation models. We also achieve competitive performance as a SOTA fully supervised method using less than one-tenth of the complete labels. Our proposed approach has the potential to improve stroke diagnosis and treatment planning, which may lead to better patient outcomes.

Semi-supervised body parsing and pose estimation for enhancing infant general movement assessment.

General movement assessment (GMA) of infant movement videos (IMVs) is an effective method for early detection of cerebral palsy (CP) in infants. We demonstrate in this paper that end-to-end trainable neural networks for image sequence recognition can be applied to achieve good results in GMA, and more importantly, augmenting raw video with infant body parsing and pose estimation information can significantly improve performance. To solve the problem of efficiently utilizing partially labeled IMVs for body parsing, we propose a semi-supervised model, termed SiamParseNet (SPN), which consists of two branches, one for intra-frame body parts segmentation and another for inter-frame label propagation. During training, the two branches are jointly trained by alternating between using input pairs of only labeled frames and input of both labeled and unlabeled frames. We also investigate training data augmentation by proposing a factorized video generative adversarial network (FVGAN) to synthesize novel labeled frames for training. FVGAN decouples foreground and background generation which allows for generating multiple labeled frames from one real labeled frame. When testing, we employ a multi-source inference mechanism, where the final result for a test frame is either obtained via the segmentation branch or via propagation from a nearby key frame. We conduct extensive experiments for body parsing using SPN on two infant movement video datasets; on these partially labeled IMVs, we show that SPN coupled with FVGAN achieves state-of-the-art performance. We further demonstrate that our proposed SPN can be easily adapted to the infant pose estimation task with superior performance. Last but not least, we explore the clinical application of our method for GMA. We collected a new clinical IMV dataset with GMA annotations, and our experiments show that our SPN models for body parsing and pose estimation trained on the first two datasets generalize well to the new clinical dataset and their results can significantly boost the convolutional recurrent neural network (CRNN) based GMA prediction performance when combined with raw video inputs.

Rapid Biomarker Screening of Alzheimer's Disease by Interpretable Machine Learning and Graphene-Assisted Raman Spectroscopy.

The study of Alzheimer's disease (AD), the most common cause of dementia, faces challenges in terms of understanding the cause, monitoring the pathogenesis, and developing early diagnoses and effective treatments. Rapid and accurate identification of AD biomarkers in the brain is critical to providing key insights into AD and facilitating the development of early diagnosis methods. In this work, we developed a platform that enables a rapid screening of AD biomarkers by employing graphene-assisted Raman spectroscopy and machine learning interpretation in AD transgenic animal brains. Specifically, we collected Raman spectra on slices of mouse brains with and without AD and used machine learning to classify AD and non-AD spectra. By contacting monolayer graphene with the brain slices, the accuracy was increased from 77% to 98% in machine learning classification. Further, using a linear support vector machine (SVM), we identified a spectral feature importance map that reveals the importance of each Raman wavenumber in classifying AD and non-AD spectra. Based on this spectral feature importance map, we identified AD biomarkers including Aß and tau proteins and other potential biomarkers, such as triolein, phosphatidylcholine, and actin, which have been confirmed by other biochemical studies. Our Raman-machine learning integrated method with interpretability will facilitate the study of AD and can be extended to other tissues and biofluids and for various other diseases.

Understanding the Excitation Wavelength Dependence and Thermal Stability of the SARS-CoV-2 Receptor-Binding Domain Using Surface-Enhanced Raman Scattering and Machine Learning.

COVID-19 has cost millions of lives worldwide. The constant mutation of SARS-CoV-2 calls for thorough research to facilitate the development of variant surveillance. In this work, we studied the fundamental properties related to the optical identification of the receptor-binding domain (RBD) of SARS-CoV-2 spike protein, a key component of viral infection. The Raman modes of the SARS-CoV-2 RBD were captured by surface-enhanced Raman spectroscopy (SERS) using gold nanoparticles (AuNPs). The observed Raman enhancement strongly depends on the excitation wavelength as a result of the aggregation of AuNPs. The characteristic Raman spectra of RBDs from SARS-CoV-2 and MERS-CoV were analyzed by principal component analysis that reveals the role of secondary structures in the SERS process, which is corroborated with the thermal stability under laser heating. We can easily distinguish the Raman spectra of two RBDs using machine learning algorithms with accuracy, precision, recall, and F1 scores all over 95%. Our work provides an in-depth understanding of the SARS-CoV-2 RBD and paves the way toward rapid analysis and discrimination of complex proteins of infectious viruses and other biomolecules.

Phase-field modeling and machine learning of electric-thermal-mechanical breakdown of polymer-based dielectrics.

Understanding the breakdown mechanisms of polymer-based dielectrics is critical to achieving high-density energy storage. Here a comprehensive phase-field model is developed to investigate the electric, thermal, and mechanical effects in the breakdown process of polymer-based dielectrics. High-throughput simulations are performed for the P(VDF-HFP)-based nanocomposites filled with nanoparticles of different properties. Machine learning is conducted on the database from the high-throughput simulations to produce an analytical expression for the breakdown strength, which is verified by targeted experimental measurements and can be used to semiquantitatively predict the breakdown strength of the P(VDF-HFP)-based nanocomposites. The present work provides fundamental insights to the breakdown mechanisms of polymer nanocomposite dielectrics and establishes a powerful theoretical framework of materials design for optimizing their breakdown strength and thus maximizing their energy storage by screening suitable nanofillers. It can potentially be extended to optimize the performances of other types of materials such as thermoelectrics and solid electrolytes.