Vision-language models (VLMs) have shown strong generalization across multimodal tasks, but adapting them to medical report generation (MRG) often demands extensive paired image-text data that are limited due to data privacy and annotation cost. In-context learning (ICL) offers a promising training-free alternative, yet standard ICL approaches rely on long demonstration prompts that are computationally inefficient and often yield inconsistent or clinically inaccurate descriptions. To address these challenges, we propose Principal In-Context Vectors (PCVs), a compact latent-guidance framework that distills multimodal demonstrations into stable semantic representations. By extracting hidden states from auto-regressive VLMs and applying principal component analysis (PCA), we identify robust semantic directions that remain stable under input perturbations. These PCVs are then injected into new queries to steer generation toward accurate and clinically meaningful outputs without any model tuning. Extensive experiments on four MRG benchmark datasets show that our approach can enhance both zero-shot and fully supervised generation quality across diverse settings, including cross-center, cross-disease, and longitudinal scenarios. This work provides a lightweight and scalable approach to adapt pre-trained VLMs for practical clinical deployment.

Understanding gene regulatory networks (GRNs) holds paramount importance for deciphering the intricate interplay among genes and their influence on biological processes and disease pathogenesis. The emergence of single-cell RNA sequencing (scRNA-seq) techniques has heralded a new era in GRN inference by capturing the nuanced heterogeneity and dynamic nature of gene expression at the single-cell level. However, extracting meaningful patterns from scRNA-seq measurements to infer GRNs poses significant challenges to existing methodologies due to the sheer scale and inherent complexity of the data. Here we propose a highly accurate and computationally efficient strategy for scRNA-seq-based GRN inference. Our approach leverages the underlying interactive relationships among the cells using state-of-the-art deep learning strategy. Specifically, a spatially semantic image representation, termed CelloGraph, is first introduced to portray the expressions of each gene across cells. The allocation of a cell to a spatial grid point of the CelloGraph is dictated by its interactions with other cells within the system, as determined by the maximization of system entropy of cell-cell interactions. Subsequently, the CelloGraphs of all pertinent genes are analyzed by using a customarily designed convolutional neural network (CNN) to discern discriminant patterns in the data and infer GRNs. The efficacy of the proposed approach is demonstrated through diverse real-world biomedical datasets. By harnessing the distinctive attributes of spatially semantic CelloGraphs and leveraging the unique pattern discovery capabilities of CNNs, our methodology paves the way for a deeper comprehension of the underlying mechanisms that govern gene expression and regulation. The proposed strategy not only overcomes challenges in scRNA-seq-based GRN inference but also promises to provide a more comprehensive understanding of intricate biological processes.

High-quality motion estimation is essential in ultrasound elastography (USE) for evaluating tissue mechanical properties and detecting abnormalities. Traditional methods, such as speckle tracking and regularized optimization, face challenges including noise, oversmoothing of displacements and prolonged runtimes. Recent efforts have explored optical flow-based convolutional neural networks (CNNs). However, current approaches face, at least, one of the following limitations: 1) reliance on tissue incompressibility assumption, which compromises data fidelity and can introduce large errors; 2) dependence on ground truth displacement data for supervised CNN methods; 3) use of a regularizer not aligned with tissue physics by relying on first-order displacement derivatives only; 4) use of a L2-norm regularizer that oversmoothes motion estimates; and 5) substantially large sampling size, increasing computational and memory demands, especially for classical methods. In this paper, we develop RAFT-USENet, a physics-informed, unsupervised neural network to estimate both axial and lateral displacements. We design RAFT-USENet by substantially modifying the optical flow RAFT network to adapt it to high-frequency USE data. Extensive validation using simulation, phantom and in vivo data demonstrates that RAFT-USENet significantly outperforms existing methods, with normalized cross-correlation values of 0.94, 0.88, and 0.82 in simulation, breast phantom and in vivo datasets, respectively, compared to 0.79-0.88, 0.76-0.85, and 0.69-0.81 for the competing methods. Additionally, RAFT-USENet reduces computational time by 1.5-150 times compared to existing methods. These results suggest that RAFT-USENet may be a potentially reliable and accurate tool for clinical elasticity imaging applications.

Single-cell RNA sequencing (scRNA-seq) has emerged as a powerful tool to gain biological insights at the cellular level. However, due to technical limitations of the existing sequencing technologies, low gene expression values are often omitted, leading to inaccurate gene counts. Existing methods, including advanced deep learning techniques, struggle to reliably impute gene expressions due to a lack of mechanisms that explicitly consider the underlying biological knowledge of the system. In reality, it has long been recognized that gene-gene interactions may serve as reflective indicators of underlying biology processes, presenting discriminative signatures of the cells. A genomic data analysis framework that is capable of leveraging the underlying gene-gene interactions is thus highly desirable and could allow for more reliable identification of distinctive patterns of the genomic data through extraction and integration of intricate biological characteristics of the genomic data. Here we tackle the problem in two steps to exploit the gene-gene interactions of the system. We first reposition the genes into a 2D grid such that their spatial configuration reflects their interactive relationships. To alleviate the need for labeled ground truth gene expression datasets, a self-supervised 2D convolutional neural network is employed to extract the contextual features of the interactions from the spatially configured genes and impute the omitted values. Extensive experiments with both simulated and experimental scRNA-seq datasets are carried out to demonstrate the superior performance of the proposed strategy against the existing imputation methods.

AI driven by deep learning is transforming many aspects of science and technology. The enormous success of deep learning stems from its unique capability of extracting essential features from Big Data for decision-making. However, the feature extraction and hidden representations in deep neural networks (DNNs) remain inexplicable, primarily because of lack of technical tools to comprehend and interrogate the feature space data. The main hurdle here is that the feature data are often noisy in nature, complex in structure, and huge in size and dimensionality, making it intractable for existing techniques to analyze the data reliably. In this work, we develop a computational framework named contrastive feature analysis (CFA) to facilitate the exploration of the DNN feature space and improve the performance of AI. By utilizing the interaction relations among the features and incorporating a novel data-driven kernel formation strategy into the feature analysis pipeline, CFA mitigates the limitations of traditional approaches and provides an urgently needed solution for the analysis of feature space data. The technique allows feature data exploration in unsupervised, semi-supervised and supervised formats to address different needs of downstream applications. The potential of CFA and its applications for pruning of neural network architectures are demonstrated using several state-of-the-art networks and well-annotated datasets across different disciplines.

Knowledge of the mechanical properties is of great clinical significance for diagnosis, prognosis and treatment of cancers. Recently, a new method based on Eshelby's theory to simultaneously assess Young's modulus (YM) and Poisson's ratio (PR) in tissues has been proposed. A significant limitation of this method is that accuracy of the reconstructed YM and PR is affected by the orientation/alignment of the tumor with the applied stress. In this paper, we propose a new method to reconstruct YM and PR in cancers that is invariant to the 3D orientation of the tumor with respect to the axis of applied stress. The novelty of the proposed method resides on the use of a tensor transformation to improve the robustness of Eshelby's theory and reconstruct YM and PR of tumors with high accuracy and in realistic experimental conditions. The method is validated using finite element simulations and controlled experiments using phantoms with known mechanical properties. The in vivo feasibility of the developed method is demonstrated in an orthotopic mouse model of breast cancer. Our results show that the proposed technique can estimate the YM and PR with overall accuracy of (97.06 ± 2.42) % under all tested tumor orientations. Animal experimental data demonstrate the potential of the proposed methodology in vivo. The proposed method can significantly expand the range of applicability of the Eshelby's theory to tumors and provide new means to accurately image and quantify mechanical parameters of cancers in clinical conditions.

Deep learning-based diagnosis is becoming an indispensable part of modern healthcare. For high-performance diagnosis, the optimal design of deep neural networks (DNNs) is a prerequisite. Despite its success in image analysis, existing supervised DNNs based on convolutional layers often suffer from their rudimentary feature exploration ability caused by the limited receptive field and biased feature extraction of conventional convolutional neural networks (CNNs), which compromises the network performance. Here, we propose a novel feature exploration network named manifold embedded multilayer perceptron (MLP) mixer (ME-Mixer), which utilizes both supervised and unsupervised features for disease diagnosis. In the proposed approach, a manifold embedding network is employed to extract class-discriminative features; then, two MLP-Mixer-based feature projectors are adopted to encode the extracted features with the global reception field. Our ME-Mixer network is quite general and can be added as a plugin to any existing CNN. Comprehensive evaluations on two medical datasets are performed. The results demonstrate that their approach greatly enhances the classification accuracy in comparison with different configurations of DNNs with acceptable computational complexity.

Semantic segmentation is an important step in understanding the scene for many practical applications such as autonomous driving. Although Deep Convolutional Neural Networks-based methods have significantly improved segmentation accuracy, small/thin objects remain challenging to segment due to convolutional and pooling operations that result in information loss, especially for small objects. This article presents a novel attention-based method called Across Feature Map Attention (AFMA) to address this challenge. It quantifies the inner-relationship between small and large objects belonging to the same category by utilizing the different feature levels of the original image. The AFMA could compensate for the loss of high-level feature information of small objects and improve the small/thin object segmentation. Our method can be used as an efficient plug-in for a wide range of existing architectures and produces much more interpretable feature representation than former studies. Extensive experiments on eight widely used segmentation methods and other existing small-object segmentation models on CamVid and Cityscapes demonstrate that our method substantially and consistently improves the segmentation of small/thin objects.

Prior studies using graph neural networks (GNNs) for image classification have focused on graphs generated from a regular grid of pixels or similar-sized superpixels. In the latter, a single target number of superpixels is defined for an entire dataset irrespective of differences across images and their intrinsic multiscale structure. On the contrary, this study investigates image classification using graphs generated from an image-specific number of multiscale superpixels. We propose WaveMesh, a new wavelet-based superpixeling algorithm, where the number and sizes of superpixels in an image are systematically computed based on its content. WaveMesh superpixel graphs are structurally different from similar-sized superpixel graphs. We use SplineCNN, a state-of-the-art network for image graph classification, to compare WaveMesh and similar-sized superpixels. Using SplineCNN, we perform extensive experiments on three benchmark datasets under three local-pooling settings: 1) no pooling, 2) GraclusPool, and 3) WavePool, a novel spatially heterogeneous pooling scheme tailored to WaveMesh superpixels. Our experiments demonstrate that SplineCNN learns from multiscale WaveMesh superpixels on-par with similar-sized superpixels. In all WaveMesh experiments, GraclusPool performs poorer than no pooling / WavePool, showing that poor cluster assignment negatively affects the performance of the network while learning from multiscale superpixels.

The automatic diagnosis of various retinal diseases from fundus images is important to support clinical decision-making. However, developing such automatic solutions is challenging due to the requirement of a large amount of human-annotated data. Recently, unsupervised/self-supervised feature learning techniques receive a lot of attention, as they do not need massive annotations. Most of the current self-supervised methods are analyzed with single imaging modality and there is no method currently utilize multi-modal images for better results. Considering that the diagnostics of various vitreoretinal diseases can greatly benefit from another imaging modality, e.g., FFA, this paper presents a novel self-supervised feature learning method by effectively exploiting multi-modal data for retinal disease diagnosis. To achieve this, we first synthesize the corresponding FFA modality and then formulate a patient feature-based softmax embedding objective. Our objective learns both modality-invariant features and patient-similarity features. Through this mechanism, the neural network captures the semantically shared information across different modalities and the apparent visual similarity between patients. We evaluate our method on two public benchmark datasets for retinal disease diagnosis. The experimental results demonstrate that our method clearly outperforms other self-supervised feature learning methods and is comparable to the supervised baseline. Our code is available at GitHub.

Novel viscoelastic and poroelastic elastography techniques rely on the accurate estimation of the temporal behavior of the axial or lateral strains and related parameters. From the temporal curve of the elastographic parameter of interest, the time constant (TC) is estimated using analytical models and curve-fitting techniques such as Levenberg-Marquardt (LM), Nelder-Mead (NM), and trust-region reflective (TR). In this paper, we propose a new technique named variable projection (VP) to estimate accurately and robustly the TC and steady-state value of the elastographic parameter of interest from its temporal curve. As a testing platform, the method is used with a novel analytical model, which can be used for both poroelastic and viscoelastic tissues and in most practical experimental conditions of clinical interest. Finite element and ultrasound simulations and experimental results demonstrate that VP is robust to noise and capable of estimating the TC of the elastographic parameter with accuracy higher than that of typically employed curve-fitting techniques. The results also demonstrate that the performance of VP is not affected by an incorrect initial TC guess. For example, in simulations, VP can estimate the TC of axial strain and effective Poisson's ratio accurately for initial guesses ranging from 0.001 to infinite times of the true TC value even in fairly noisy conditions (30-dB signal to noise ratio). In experiments, VP always estimates the axial strain TC reliably, whereas the LM, NM, and TR methods fail to converge or converge to wrong solutions in most of the cases.

Ultrasound poroelastography aims at assessing the poroelastic behavior of biological tissues via estimation of the local temporal axial strains and effective Poisson's ratios (EPR). Currently, reliable estimation of EPR using ultrasound is a challenging task due to the limited quality of lateral strain estimation. In this paper, we propose a new two-step EPR estimation technique based on dynamic programming elastography (DPE) and Horn-Schunck (HS) optical flow estimation. In the proposed method, DPE is used to estimate the integer axial and lateral displacements while HS is used to obtain subsample axial and lateral displacements from the motion-compensated pre-compressed and post-compressed radio frequency data. Axial and lateral strains are then calculated using Kalman filter-based least square estimation. The proposed two-step technique was tested using finite-element simulations, controlled experiments and in vivo experiments, and its performance was statistically compared with that of analytic minimization (AM) and correlation-based method (CM). Our results indicate that our technique provides EPR elastograms of higher quality and accuracy than those produced by AM and CM. Regarding signal-to-noise ratio and elastographic contrast-to-noise ratio, in simulated data, the proposed method provides an average improvement of 30% and 75%, respectively, with respect to AM and of 100% and 169%, respectively, with respect to CM, whereas, in experiments, the proposed approach provides an average improvement of 30% and 67% with respect to AM and of 230% and 525% with respect to CM. Based on these results, the proposed method may be the preferred one in experimental poroelastography applications.

The mechanical behavior of long bones and fractures has been under investigation for many decades due to its complexity and clinical relevance. In this paper, we report a new subject-specific methodology to predict and analyze the mechanical behavior of the soft tissue at a bone interface with the intent of identifying the presence and location of bone abnormalities with high accuracy, spatial resolution, and contrast. The proposed methodology was tested on both intact and fractured rabbit femur samples with finite element-based 3-D simulations, created from actual femur computed tomography data, and ultrasound elastography experiments. The results included in this study demonstrate that elastographic strains at the bone/soft tissue interface can be used to differentiate fractured femurs from the intact ones on a distribution level. These results also demonstrate that coronal plane axial shear strain creates a unique contrast mechanism that can be used to reliably detect fractures (both complete and incomplete) in long bones. Kruskal-Wallis test further demonstrates that the contrast measure for the fracture group (simulation: 2.1286±0.2206; experiment: 2.7034 ± 1.0672) is significantly different from that for the intact group (simulation: 0 ± 0; experiment: 1.1540±0.6909) when using coronal plane axial shear strain elastography ( < 0.01). We conclude that: 1) elastography techniques can be used to accurately identify the presence and location of fractures in a long bone and 2) the proposed model-based approach can be used to predict and analyze strains at a bone fracture site and to better interpret experimental elastographic data.

This paper presents a speech enhancement approach, where an adaptive threshold is statistically determined based on Student$t$Modeling of Teager energy (TE) operated perceptual wavelet packet (PWP) coefficients of noisy speech. In order to obtain an enhanced speech, the threshold thus derived is applied upon the PWP coefficients by employing a Student$t$pdf dependent custom thresholding function, which is designed based on a combination of modified hard and semisoft thresholding functions. Extensive simulations are carried out using the NOIZEUS database to evaluate the effectiveness of the proposed method for car and multi-talker babble noise corrupted speech signals. Several standard objective measures and subjective evaluations including formal listening tests show that the proposed method outperforms some of the state-of-the-art speech enhancement methods at high as well as low levels of SNRs.