Number Of Layers Research Articles

Adaptive Evolutionary Optimization of Deep Learning Architectures for Focused Liver Ultrasound Image Segmentation.

Background: Liver ultrasound segmentation is challenging due to low image quality and variability. While deep learning (DL) models have been widely applied for medical segmentation, generic pre-configured models may not meet the specific requirements for targeted areas in liver ultrasound. Quantitative ultrasound (QUS) is emerging as a promising tool for liver fat measurement; however, accurately segmenting regions of interest within liver ultrasound images remains a challenge. Methods: We introduce a generalizable framework using an adaptive evolutionary genetic algorithm to optimize deep learning models, specifically U-Net, for focused liver segmentation. The algorithm simultaneously adjusts the depth (number of layers) and width (neurons per layer) of the network, dropout, and skip connections. Various architecture configurations are evaluated based on segmentation performance to find the optimal model for liver ultrasound images. Results: The model with a depth of 4 and filter sizes of [16, 64, 128, 256] achieved the highest mean adjusted Dice score of 0.921, outperforming the other configurations, using three-fold cross-validation with early stoppage. Conclusions: Adaptive evolutionary optimization enhances the deep learning architecture for liver ultrasound segmentation. Future work may extend this optimization to other imaging modalities and deep learning architectures.

Effect of Depth of Cut and Number of Layers on the Surface Roughness and Surface Homogeneity After Milling of Al/CFRP Stacks.

A multilayer structure is a type of construction consisting of outer layers and a core, which is mainly characterized by high strength and specific stiffness, as well as the ability to dampen vibration and sound. This structure combines the high strength of traditional materials (mainly metals) and composites. Currently, sandwich structures in any configurations (types of core) are one of the main directions of technology development and research. This paper evaluates the surface quality of II- and III-layer sandwich structures that are a combination of aluminum alloy and CFRP (Carbon Fiber-Reinforced Polymer) after the machining. The effect of depth of cut (ae) on the surface roughness of the II- and III-layer sandwich structures after the milling process was investigated. The surface homogeneity was also investigated. It was expressed by the IRa and IRz surface homogeneity indices formed from the Ra and Rz surface roughness parameters measured separately for each layer of the materials forming the sandwich structure. It was noted that the lowest surface roughness (Ra = 0.03 µm and Rz = 0.20 µm) was obtained after the milling of the II-layer sandwich structure using ae = 0.5 mm, while the highest was obtained for the III-layer structure and ae = 1.0 mm (Ra = 1.73 µm) and ae = 0.5 mm (Rz = 10.98 µm). The most homogeneous surfaces were observed after machining of the II-layer structure and using the depth of cut ae = 2.0 mm (IRa = 0.28 and IRz = 0.06), while the least homogeneous surfaces were obtained after milling of the III-layer structure and the depths of cut ae = 0.5 mm (IRa = 0.64) and ae = 2.0 mm (IRz = 0.78). The obtained results may be relevant to surface engineering and combining hybrid sandwich structures with other materials.

Enhancing Underwater Object Recognition: Integrating Transfer Learning with Hybrid Optimization Techniques for Improved Detection Accuracy

Underwater object recognition presents unique challenges due to varying water conditions, low visibility, and the presence of noise. This research proposes an advanced methodology that combines transfer learning and hybrid optimization techniques to enhance recognition accuracy in underwater environments. Specifically, a pre-trained EfficientNet model is employed for feature extraction, leveraging its capacity to capture diverse features in underwater images. The model is then optimized using a hybrid Particle Swarm Optimization and Genetic Algorithm (PSOGA) to fine-tune hyperparameters such as learning rate, number of layers, and activation functions. This hybrid approach balances exploration and exploitation in the search space, allowing the model to converge on an optimal solution that maximizes accuracy. The model is evaluated against nine existing deep learning models, including ResNet-50, VGG-16, EfficientNet-B0, and MobileNetV2. The proposed PSOGA model achieves a superior accuracy of 98.32%, surpassing the best-performing models like EfficientNet-B0, which reached 95.89%. Furthermore, the model outperforms traditional optimizers like Adam, RMSprop, and AdaGrad, which attained lower accuracies. Precision, recall, and F1-score for the PSOGA model also demonstrate remarkable improvements, highlighting the model's effectiveness in underwater object recognition. The combination of transfer learning and hybrid optimization enables the model to generalize well across diverse underwater environments while maintaining computational efficiency.

Elementary-level intrusive coupling of machine learning for efficient mechanical analysis of variable stiffness composite laminates: a spatially-adaptive fidelity-sensitive computational framework

AbstractMechanical analysis of the complex configurations of composite laminates can be computationally prohibitive based on accurate higher-order theories, especially when the analyses involve multiple realizations corresponding to different sets of input parameters such as uncertainty quantification, optimization, reliability and sensitivity analysis. Efficient lower-order theories should not be adopted in such situations since the error accumulates with multiple realizations, leading to poor outcomes. We propose an elementary-level coupling of machine learning for efficient, yet accurate mechanical analysis of laminated composites based on finite element simulations coupled with gaussian process regression. The generic parameter space of material properties, mesh size, number of layers, and ply angle in composite laminates are accounted for forming an efficient mapping with the augmentation of lower-order theory-based elementary-level structural matrices. The computationally efficient machine learning models predict the difference in the elements of the stiffness matrix for higher-order zigzag theory (HOZT) and first-order shear deformation theory (FSDT) at the first stage. Based on such machine learning-based difference mapping, we augment the elementary stiffness matrices obtained using FSDT efficiently to the equivalent of HOZT theory without any additional computational expenses (referred to here as augmented FSDT, or aFSDT). However, it is not necessary to augment all the elements in the analysis domain which might otherwise lead to unnecessary computational expenses and loss in accuracy. To achieve an optimal level of computational efficiency and accuracy, we further propose spatially-adaptive fidelity-sensitive coupling of machine learning, only for the elements within the analysis domain where it is necessary to adopt higher-order theories. The selective augmentation strategy essentially brings in a scope of integrating physics-based insights of critical stress resultant distribution into the algorithm based on best theory diagram. Subsequently, the global structural matrices are computed exploiting such adaptive criteria containing a mixed set of elements formed using FSDT and aFSDT, which leads to an accuracy equivalent to HOZT in the mechanical analysis of composite laminates almost at the computational expense of FSDT. The proposed spatially-adaptive fidelity-sensitive scheme ensures optimal performance in terms of computational efficiency by augmenting selective elements while minimizing the loss of accuracy due to the involvement of surrogates. Detailed numerical results are presented for static, dynamic and stability characterization of composite laminates including the demonstration for variable stiffness composite configurations based on the efficient machine learning-assisted elementary-level intrusive computational framework, wherein the notion of engineering judgement is introduced concerning the trade-off between computational efficiency and required level of accuracy.

A COMPUTER PROGRAM FOR SLABS WITH DISCONTINUITIES ON LAYERED ELASTIC SOLIDS

A finite element method programmed for a high­ speed computer is presented. The method is based on the classical theory of thin plate. The program is developed for subgrade soil represented as a linear layered elastic solid. Any number of layers can be accommodated. The program is capable of analyzing stress conditions in concrete pavements with the load transfer in the joint, as well as in the supporting subgrade soil, subjected to loads and temperature warping. Multiple-wheel loads can be input and the number of wheels is not limited. Because of the large computer storage space required, the program can handle only two slabs, except for the special option where a four-slab pavement system is loaded symmetrically at the pavement's center. At the joint, the progam considers only the shear transfer and assumes the element transfer to be zero. The results computed using the program were compared with stress transfers across the joint measured in airfield test pavements. The comparisons were very favorable.

Benchmarking the speed-accuracy tradeoff in object recognition by humans and neural networks.

Active object recognition, fundamental to tasks like reading and driving, relies on the ability to make time-sensitive decisions. People exhibit a flexible tradeoff between speed and accuracy, a crucial human skill. However, current computational models struggle to incorporate time. To address this gap, we present the first dataset (with 148 observers) exploring the speed-accuracy tradeoff (SAT) in ImageNet object recognition. Participants performed a 16-way ImageNet categorization task where their responses counted only if they occurred near the time of a fixed-delay beep. Each block of trials allowed one reaction time. As expected, human accuracy increases with reaction time. We compare human performance with that of dynamic neural networks that adapt their computation to the available inference time. Time is a scarce resource for human object recognition, and finding an appropriate analog in neural networks is challenging. Networks can repeat operations by using layers, recurrent cycles, or early exits. We use the repetition count as a network's analog for time. In our analysis, the number of layers, recurrent cycles, and early exits correlates strongly with floating-point operations, making them suitable time analogs. Comparing networks and humans on SAT-fit error, category-wise correlation, and SAT-curve steepness, we find cascaded dynamic neural networks most promising in modeling human speed and accuracy. Surprisingly, convolutional recurrent networks, typically favored in human object recognition modeling, perform the worst on our benchmark.

Optimal sandwich panel's core design for an enhanced impact resistance.

Despite the extensive literature revealing various core structures that can enhance the impact resistance of composite panels, a comparative study illustrating the difference in performance of the various cores under same loading conditions is missing. The aim of this study is to determine the optimal core structure and design in terms of energy absorption under low-velocity impact using both numerical simulations and experimental testing for validation. Response surface analysis was used to design the experiments and analyse the panel's behaviour. A total of 160 numerical simulations were conducted by varying the core shape, density, number of layers and panel's thickness. Drop tower tests were performed to experimentally validate the results. Additive manufacturing was used to 3D print the tested structures which simplified the manufacturing process. Results provided insights on time to perforation, stiffness of the panels showcasing an inverse relationship between recoverable strain energy and equivalent plastic strain. The number of layers within the panels were identified as a pivotal factor in the energy distribution and tendency for localized plastic deformation to occur. Both numerical and experimental results revealed the superior energy absorption capabilities of the X-frame core shaped structure. Regression models developed shed light on the relationships between core height, volume fraction, number of layers, and core topology revealing the extent of damage. Multi-objective optimization was used to yield optimal configuration for sandwich panels with highest impact resistance.

Ultrasonographic Assessment of Uterine Measurements and Endometrial Thickness Among Healthy Saudi Females Sample.

The present study was conducted to analyze uterine measurements and endometrial thickness throughout the menstrual cycle in Saudi healthy females of reproductive age. This cohort study was conducted at Princess Nourah bint Abdulrahman University, Saudi Arabia, among thirty-three females of reproductive age who underwent trans-abdominal pelvic ultrasound scans across four menstrual cycle phases. Data analysis was conducted using SPSS version 26, utilizing descriptive statistics, one-way ANOVA, correlation, and regression analysis. Endometrial thickness and layers showed significant variations (p<0.001) across menstrual phases (early proliferative: 0.59 ± 0.21 cm, late proliferative: 0.77 ± 0.24 cm, secretory: 1.09 ± 0.40 cm, menstrual: 0.52 ± 0.35 cm). Endometrial thickness was positively correlated with number of layers (r=0.576, p<0.05). The study showed that the average uterine length, width, and thickness were 7.33 ± 0.76 cm, 3.93 ± 1.00 cm and 3.44 ± 0.55 cm, which showed stability across menstrual phases, except for width showing slight variations. Endometrial thickness was positively correlated with uterine thickness (r=0.358, p<0.05). The study results emphasize the significance of using region-specific reference values in clinical practice. This approach would enable precise evaluation and treatment of gynecological problems. It is encouraged to do future study with larger populations in order to validate these results and improve the therapeutic applicability.

Exploring deformability in 3D tufted composite reinforcements: Understanding bending behaviors in forming applications

In simulations, the bending stiffness of fibrous reinforcements plays a crucial role in accurately predicting the morphology of wrinkle defects during the forming process. Three-dimensional (3D) tufted reinforcements exhibit distinct bending behaviors compared to traditional two-dimensional (2D) reinforcements due to the presence of tufting yarns in the thickness direction. This study employs cantilever bending tests to measure the bending stiffness of tufted reinforcements, examining the effects of various parameters, including tufting points, tufting distribution, number of layers, and tufting loop length, on their bending behavior. The experimental results show that through-thickness tufting yarns significantly enhance the bending stiffness of multilayered reinforcements, but have little impact on single-layer reinforcements. The improvement in bending stiffness is nonlinearly related to the number of tufting points, fabric layers, and tufting loop length, while tufting distribution has no effect. The underlying mechanisms of these tufting parameters are analyzed, and an analytical model is developed to explain the influence of tufting loop length.

Numerical and experimental analysis of body armor polymer penetration resistance against 7.62mm bullet.

Advanced materials are crucial for enhancing soldier safety through improved personal body armor. In contrast to conventional Kevlar-epoxy composites, this study examines the ballistic performance of a unique ECO-UHMWPE (Ultra-High Molecular Weight Polyethylene) vest. The aim is to achieve a lightweight design with superior impact resistance, addressing limitations of the current armor used by the Ethiopian Defense Force. A comprehensive finite element analysis (FEA) was conducted using Abaqus explicit dynamics software to simulate the impact of 7.62×39 mm projectiles at various ranges (50m, 100m, 150m, and 200m). Material properties, laminate thickness, projectile velocity, and number of layers were systematically varied to assess their impact on vest performance. Results showed that ECO-UHMWPE vests exhibited significantly lower stress values and minimal deformation, particularly at close ranges with high projectile kinetic energy, outperforming Kevlar vests. Ballistic testing validated the FEA, with ECO-UHMWPE vests achieving a 22.44% weight reduction (3.49kg, 12mm thick) compared to traditional Kevlar vests (4.5kg, 15mm thick) and offering a 30% reduction in deformation and a 25% increase in energy absorption. This analysis demonstrates the potential of ECO-UHMWPE for lighter, more effective body armor, paving the way for next-generation personal protective equipment.

Electromagnetic shielding effectiveness of three-dimensional multilayered interlaced woven fabrics using stainless steel fibers.

This study explores and discusses the design, the manufacturing and the morphology of three-dimensional (3D) multilayered weft interlaced woven fabrics using stainless steel fibers on the electromagnetic shielding efficiency (SE). Design solutions of 3D multilayered interlaced fabrics in relation to electromagnetic shielding efficiency are still not sufficiently investigated. Moreover, this study aims to analyze the differences in the internal geometry of 3D multilayered weft interlaced fabrics with different number of layers and frequency of connecting points in multilayered woven fabrics on electromagnetic SE. For this study, the input staple yarn 2×20 tex (mixed 80% polyester/20% stainless steel fibers) is used for production of approximately 28 types of different 3D multilayer weft interlaced woven fabrics (two and three-layer), with connecting points ranging from 0.5cm×0.5cm-5cm×5cm. The comparison of internal fabric microstructures indicates statistically significant differences in relation to SE, and this contribution extends the theoretical guidance of woven fabric construction with the design and production of special 3D multilayered interlaced fabrics with controlled SE.

Investigation and comparative analysis of polysilicon and MoS2-based channels in 3D NAND

Abstract In 3D NAND, channel current sensing deteriorates as the number of stacked layers increases due to the granular nature of the polysilicon channel. Recent studies on 2D materials like MoS2 have demonstrated its exceptional channel material properties, including high mobility and no interfacial dangling bonds. We incorporate MoS2 as the channel material in 3D NAND, demonstrating better cell current and lower subthreshold slope than conventional polysilicon channels to overcome the low channel current in 3D NAND. The scaling performance of the MoS2 channel device shows the gate and spacer lengths can be scaled down to 10 nm without degrading the program and erasing performance. Additionally, MoS2-based channel NAND suppresses WL interference by reducing threshold voltage shifts of the cells adjacent to the program cells. These results show MoS2 as a promising alternative channel material for 3D NAND, offering enhanced bit density growth and overall performance in flash memory applications.

Prediction of melt pool morphology of Ti alloy based on deep learning method

This paper utilizes an artificial neural network (ANN) model to predict the melt pool morphology (including melt pool width, height and depth) of Titanium (Ti) alloy by adjusting welding process parameters (laser power, welding current, and welding speed). The correlation between welding process parameters and melt pool morphology are analyzed using the Pearson correlation coefficient, which the result indicates that the input features are directly related to the output features. Four deep learning (DL) models are evaluated, which ANN model has higher coefficient of determination (0.8725 and 0.9691), lower root mean square error (RMSE) (0.4322mm and 0.1405mm) and lower mean absolute error (MAE) (0.025mm and 0.112mm). When the number of hidden layers is five, the R2 and RMSE values of ANN model are 0.8024, 0.9838, 0.5356mm and 0.1019mm, respectively. To assess the accuracy of the model, experiments are conducted to validate the predicted values, and the errors between the experimental and predicted values are 0.01mm, 0.10mm, and 0.02mm, respectively.

Kinetic multilayer models for surface chemistry in indoor environments.

Multiphase interactions and chemical reactions at indoor surfaces are of particular importance due to their impact on air quality in indoor environments with high surface to volume ratios. Kinetic multilayer models are a powerful tool to simulate various gas-surface interactions including partitioning, diffusion and multiphase chemistry of indoor compounds by treating mass transport and chemical reactions in a number of model layers in the gas and condensed phases with a flux-based approach. We have developed a series of kinetic multilayer models that have been applied to describe multiphase chemistry and interactions indoors. They include the K2-SURF model treating the reversible adsorption of volatile organic compounds on surfaces, the KM-BL model treating diffusion through an indoor surface boundary layer, the KM-FILM model treating organic film formation by multi-layer adsorption and film growth by absorption of indoor compounds, and the KM-SUB-Skin-Clothing model treating reactions of ozone with skin lipids in skin and clothing. We also developed the effective mass accommodation coefficient that can treat surface partitioning by effectively taking into account kinetic limitations of bulk diffusion. In this study we provide detailed instructions and code annotations of these models for the model user. Example sensitivity simulations that investigate the impact of input parameters are presented to help with familiarization to the codes. The user can adapt the codes as required to model experimental and indoor field campaign measurements, can use the codes to gain insights into important reactions and processes, and can extrapolate to new conditions that may not be accessible by measurements.

Modeling dispersion of circumferential waves in underwater targets with spectral methods.

The dispersion of circumferential waves propagating around cylindrical and spherical underwater targets with an arbitrary number of elastic and fluid layers is modeled using the spectral collocation method. The underlying differential equations are discretized by Chebyshev interpolation and the corresponding differentiation matrices, and the calculation of the dispersion curves is transformed into a generalized eigenvalue problem. Furthermore, for targets in infinite fluid, the perfect matched layer is used to emulate the Sommerfeld radiation condition. For solid targets, a transformation of potential functions, along with the corresponding boundary condition, is introduced to eliminate the singularity of the low-order modes at the origin. Numerical results are presented and compared with results obtained by the winding number integral method to verify the accuracy and efficiency of the approach.

Fracture propagation characteristics and energy evolution law of deep coal seam containing gangue based on discrete lattice method

Hydraulic fracturing is a key technology for the development of deep coal-rock reservoirs. Gangue, a common rock type within coal-bearing strata, significantly influences fracture morphology after fracturing. Understanding fracture propagation in coal-rock containing gangue and effectively controlling artificial fractures remain critical challenges in fracturing operations. To address this, the study investigates the coal-bearing strata of the Benxi Formation in the Daning-Jixian area of the Ordos Basin, characterized by the presence of gangue. A three-dimensional discrete lattice model was developed to analyze the effects of gangue-related factors, including the number, thickness, dip angle, and type, on hydraulic fracturing. The fracture propagation patterns and stimulation effects under these factors were systematically compared. Using node path tracking and post-processing techniques, the study identified the fracturing characteristics and fracture propagation modes at different stages. The findings indicate that gangue characteristics significantly impact fracture morphology and propagation paths. Specifically, an increase in the number of gangue layers enhances fracture complexity but reduces the effective stimulation ratio. Gangue thickness positively correlates with fracture tortuosity and inversely correlates with the effective stimulation ratio. The dip angle of gangue determines the direction of fracture propagation but has minimal influence on the stimulation area. The gangue type also affects fracture morphology; for instance, sandstone gangue leads to narrower fractures and more interface fractures at the sandstone–coal boundary compared to mudstone gangue. Fracture propagation, characterized by energy changes, can be divided into four distinct stages: Stage I (initial injection phase), Stage II (onset of fracture propagation), Stage III (fluid energy storage within gangue), and Stage IV (fracture penetration through gangue). These stages are marked by variations in fluid injection energy, fluctuation energy, surface energy, and strain energy as the fracture penetrates the layer, deflects, and forms horizontal fractures. The simulation of fracture network evolution in coal-bearing strata containing gangue provides significant theoretical guidance for understanding, predicting, and controlling fracture network morphology in coal reservoirs. These findings are instrumental for the efficient development of deep coalbed methane.

Remaining useful life prediction method of bearings based on the interactive learning strategy

To address the issues of low accuracy, high time cost, and different failure modes in predicting the remaining useful life of rolling bearings, a remaining useful life prediction model based on an interactive learning convolution network with a feature attention mechanism (ILCANet) was proposed in this paper. The model divided time series data into an odd part and an even part, and the interactive learning strategy improved the ability of the model to extract features from long series. The binary tree structure was used to increase the number of network layers and to subdivide sequence features. In the model, residual connection was introduced to prevent the gradient from disappearing. In addition, in order to determine the degree of contribution of different features, a feature attention mechanism was embedded to assign weights to features. Experiments were conducted with PHM2012 and XJTU-SY datasets, and the results showed that the proposed method had better prediction accuracy in RUL prediction than other prediction methods.

Lack of intracranial atherosclerosis in various atherosclerotic mouse models.

Although mice are used extensively to study atherosclerosis of different vascular beds, limited data is published on the occurrence of intracranial atherosclerosis. Since intracranial atherosclerosis is a common cause of stroke and is associated with dementia, a relevant animal model is needed to study these diseases. We examined the presence of intracranial atherosclerosis in different atherogenic mouse strains and studied differences in vessel wall characteristics in mouse and human tissue in search for possible explanations for the different atherosclerotic susceptibility between extracranial and intracranial vessels. The presence of atherosclerotic plaques was systematically examined from the distal common carotids to the circle of Willis in three atherogenic mouse models. Extra- and intracranial vessel characteristics were studied by immunohistochemistry. All three strains developed atherosclerotic lesions in the common carotids, while no lesions were found intracranially. This coincided with altered vessel morphology. Compared to extracranial sections, intracranially the number of elastic layers decreased, tight junction markers increased and antioxidant enzyme heme oxygenase (HO)-1 increased. Higher HO-1 expression was also shown in human intracranial arteries. Human brain endothelial cell stimulation with oxLDL induced endogenous protective antioxidant HO-1 levels through Nrf2 translocation. Intracranial atherosclerosis was absent in three atherogenic mouse models. Intracranial vessel segments showed increased presence of junction markers in mice and increased HO-1 in both mice and human tissue. We suggest that differences in brain vessel structure and induced antioxidant levels in the brain endothelium found in mouse and human tissue may contribute to decreased atherosclerosis susceptibility of intracranial arteries.

Experimental Study on the Axial Compression Performance and Bearing Capacity Calculation Method of Concrete Filled CFRP-PVC Square Tube Column

Abstract To explore the mechanical characteristics of carbon fiber reinforced polymer–polyvinyl chloride (CFRP-PVC) square tube confined concrete columns, a set of nine specimens was meticulously crafted, varying in parameters such as the quantity of CFRP layers as well as the width and spacing of CFRP strips. Axial compression tests were conducted to gauge the response of the specimens under stress. Detailed observation of the complete failure process yielded crucial parameters including axial load displacement, peak bearing capacity, ductility coefficient, and axial compression stiffness. The results unveiled distinct failure patterns: specimens under strong restraint exhibited tearing failure at the corners of the CFRP-PVC tube, whereas those with weaker restraint experienced buckling failure of the PVC tube. Moreover, adding a greater number of CFRP layers led to a substantial 46.7 % growth in maximum peak bearing capacity, albeit at the cost of a 34.5 % reduction in ductility. Similarly, widening the CFRP strip resulted in a significant 41.8 % boost in maximum peak bearing capacity and a 9.6 % growth in ductility, surpassing those of specimens lacking CFRP. Remarkably, among specimens with equivalent CFRP content, those featuring CFRP strips spaced at 50 mm and with a width of 50 mm showcased superior mechanical properties. The finite element analysis adeptly replicated the observed cracking behavior. Drawing from the experimental findings and established literature calculation methods, a robust method for determining the axial compressive bearing capacity of CFRP-PVC square tube confined concrete columns is proposed.

UnICLAM: Contrastive representation learning with adversarial masking for unified and interpretable Medical Vision Question Answering.

Medical Visual Question Answering aims to assist doctors in decision-making when answering clinical questions regarding radiology images. Nevertheless, current models learn cross-modal representations through residing vision and text encoders in dual separate spaces, which inevitably leads to indirect semantic alignment. In this paper, we propose UnICLAM, a Unified and Interpretable Medical-VQA model through Contrastive Representation Learning with Adversarial Masking. To achieve the learning of an aligned image-text representation, we first establish a unified dual-stream pre-training structure with the gradually soft-parameter sharing strategy for alignment. Specifically, the proposed strategy learns a constraint for the vision and text encoders to be close in the same space, which is gradually loosened as the number of layers increases, so as to narrow the distance between the two different modalities. For grasping the unified semantic cross-modal representation, we extend the adversarial masking data augmentation to the contrastive representation learning of vision and text in a unified manner. While the encoder training minimizes the distance between the original and masking samples, the adversarial masking module keeps adversarial learning to conversely maximize the distance. We also intuitively take a further exploration of the unified adversarial masking augmentation method, which improves the potential ante-hoc interpretability with remarkable performance and efficiency. Experimental results on VQA-RAD and SLAKE benchmarks demonstrate that UnICLAM outperforms existing 11 state-of-the-art Medical-VQA methods. More importantly, we make an additional discussion about the performance of UnICLAM in diagnosing heart failure, verifying that UnICLAM exhibits superior few-shot adaption performance in practical disease diagnosis. The codes and models will be released upon the acceptance of the paper.