Evolutionary algorithms serve as a pivotal tool in addressing black-box problems, finding widespread applications across diverse academic disciplines and engineering domains. Despite their utility, these algorithms often confront challenges when navigating complex search spaces, impeding a comprehensive exploration of potential solutions. Solely depending on the algorithm’s exploration abilities falls short of fully harnessing the rich information contained within search spaces. To unlock the full potential of solution spaces, we introduce a deep learning method for the reconstruction of the search path. Specifically, discrete data sampled by evolutionary operators during the exploration process are collected, and a uniquely designed fully connected neural network is employed to reconstruct the exploration paths. The neural network’s robust fitting capability facilitates the transformation of initially discrete sampled information into a continuous form. By capitalizing on the reconstructed solution space information, the algorithm excels in identifying superior solutions. We refer to this method of deep learning-based search path reconstruction evolution strategy algorithm (DLES). The effectiveness of DLES is validated across multiple datasets, including CEC 2014, CEC 2018, CEC 2022 and BBOB. Experimental results, compared to several state-of-the-art algorithms, affirm the superiority of the DLES algorithm.

Medical image segmentation serves as an important tool in the treatment of various medical diseases. However, achieving precise and efficient segmentation remains challenging due to the intricate structures and variations. Although neural network methods based on U-shaped structures have shown impressive results, they often lack effective representation of global–local features, leading to insufficient extraction of multi-scale and contextual information in medical image segmentation tasks. To tackle these challenges, we propose a novel approach: a dilated dendritic module with deep supervision, namely 3DL-Net. It integrates the flexible dilated convolution mechanism into the segmentation architecture, aiming to expand the model’s receptive field and capture richer global features. Additionally, in contrast to other segmentation architectures, we innovatively introduce the processing of local feature of shallow structures through the dendritic neuron module in medical images into 3DL-Net. This is the first time dendritic learning has been employed at the channel level and represents a pioneering approach to the local feature process. During the training process, 3DL-Net incorporates a deep supervision mechanism that utilizes our designed loss function. Due to the intermediate supervision signals at various network stages, providing feedback at multiple levels, the model refines its predictions across various scales, contributing to further enhanced segmentation outcomes. To evaluate the effectiveness of our proposed method, we conducted extensive experiments on three medical image datasets to demonstrate significant improvements in segmentation accuracy compared to state-of-the-art models. Our mDice metrics on three datasets achieved 86.61%, 87.87%, and 85.06%, surpassing the second-best models by 3.85%, 1.54%, and 0.95%. • We propose 3DL-Net for medical image segmentation. • We proposed a dilated conv-based module precise global features representation. • We proposed a dendritic learning-based module refine local features process. • Deep supervision & customized loss optimize training, capture fine-grained details. • Proposed method outperforms existing networks on multiple datasets.

Ultrasound detection is a potent tool for the clinical diagnosis of various diseases due to its real-time, convenient, and noninvasive qualities. Yet, existing ultrasound beamforming and related methods face a big challenge to improve both the quality and speed of imaging for the required clinical applications. The most notable characteristic of ultrasound signal data is its spatial and temporal features. Because most signals are complex-valued, directly processing them by using real-valued networks leads to phase distortion and inaccurate output. In this study, for the first time, we propose a complex-valued convolutional gated recurrent (CCGR) neural network to handle ultrasound analytic signals with the aforementioned properties. The complex-valued network operations proposed in this study improve the beamforming accuracy of complex-valued ultrasound signals over traditional real-valued methods. Further, the proposed deep integration of convolution and recurrent neural networks makes a great contribution to extracting rich and informative ultrasound signal features. Our experimental results reveal its outstanding imaging quality over existing state-of-the-art methods. More significantly, its ultrafast processing speed of only 0.07 s per image promises considerable clinical application potential. The code is available at https://github.com/zhangzm0128/CCGR.

Breast tumor segmentation in ultrasound images remains a challenging task due to low contrast, acoustic shadowing, and heterogeneous tumor appearance. Traditional deep learning-based segmentation models often perform poorly in addressing these challenges, making it difficult to accurately capture fine-grained tumor boundaries and complex structural variations. To address these issues, we propose a novel component—dendritic kernel convolution, inspired by the synaptic integration and inhibition mechanisms of biological neurons. Unlike traditional convolutional kernels that perform only linear weighting operations, dendritic kernel convolution simulates the nonlinear excitation and inhibition mechanisms of dendritic computation, adjusting feature aggregation and boundary optimization strategies. This mechanism enhances key information while suppressing noise, effectively reducing missed detections and erroneous segmentations, thereby improving segmentation accuracy and robustness. Inspired by the hierarchical processing mechanism of the human visual system—which progresses from coarse to fine perception—we further design a multistage refinement architecture. Based on dendritic kernel convolution, we construct two key modules: adendritic-dilated convolution moduleand adendritic U-Net module, and integrate them into a unified framework, termed the dendritic kernel convolutional neural network (DKNet) for breast tumor segmentation. To assess the segmentation performance of the proposed network, we conduct a comparative analysis against several state-of-the-art segmentation methods using seven quantitative metrics. The experimental results unequivocally demonstrate that DKNet surpasses all other methods, exhibiting superior segmentation outcomes and affirming its efficacy for breast tumor segmentation.

Feature selection (FS) has garnered significant attention because of its pivotal role in enhancing the efficiency and effectiveness of various machine learning and data mining algorithms. Concurrently, multiobjective feature selection (MOFS) algorithms strive to balance the complexity of multiple optimization objectives during the FS process. These include minimizing the number of selected features while maximizing classification performance. Nonetheless, managing the complexity of feature combinations presents a formidable challenge, particularly in high-dimensional datasets. Evolutionary algorithms (EAs) are increasingly adopted in MOFS owing to their exceptional global search capabilities and robustness. Despite their strengths, EAs face difficulties in navigating expansive solution spaces and achieving a balance between exploration and exploitation. To address these challenges, this study introduces a novel information gain-based EA for MOFS, designated as IGEA. This approach utilizes a clustering method for selecting a diverse parent population, thereby enhancing individual variability and maintaining a high-quality population. Considerably, IGEA employs information gain as a metric to evaluate the contribution of features to classification tasks. This metric informs crucial operations such as crossover and mutation. Moreover, the study extensively examines the actual solutions derived from IGEA, focusing on feature correlation and redundancy. This analysis illuminates IGEA's adept handling of these aspects to refine MOFS. Experimental results on 23 widely used classification datasets confirm IGEA's superiority over five other state-of-the-art algorithms, demonstrating its enhanced effectiveness and efficiency in complex MOFS scenarios.

Ultrasound imaging is widely used in medical diagnosis. It has the advantages of being performed in real time, cost-efficient, noninvasive, and nonionizing. The traditional delay-and-sum (DAS) beamformer has low resolution and contrast. Several adaptive beamformers (ABFs) have been proposed to improve them. Although they improve image quality, they incur high computation cost because of the dependence on data at the expense of real-time performance. Deep-learning methods have been successful in many areas. They train an ultrasound imaging model that can be used to quickly handle ultrasound signals and construct images. Real-valued radio-frequency signals are typically used to train a model, whereas complex-valued ultrasound signals with complex weights enable the fine-tuning of time delay for enhancing image quality. This work, for the first time, proposes a fully complex-valued gated recurrent neural network to train an ultrasound imaging model for improving ultrasound image quality. The model considers the time attributes of ultrasound signals and uses complete complex-number calculation. The model parameter and architecture are analyzed to select the best setup. The effectiveness of complex batch normalization is evaluated in training the model. The effect of analytic signals and complex weights is analyzed, and the results verify that analytic signals with complex weights enhance the model performance to reconstruct high-quality ultrasound images. The proposed model is finally compared with seven state-of-the-art methods. Experimental results reveal its great performance.

Protein structure prediction (PSP) is predicting the three-dimensional of protein from its amino acid sequence only based on the information hidden in the protein sequence. One of the efficient tools to describe this information is protein energy functions. Despite the advancements in biology and computer science, PSP is still a challenging problem due to its large protein conformation space and inaccurate energy functions. In this study, PSP is treated as a many-objective optimization problem and four conflicting energy functions are used as different objectives to be optimized. A novel Pareto-dominance-archive and Coordinated-selection-strategy-based Many-objective-optimizer (PCM) is proposed to perform the conformation search. In it, convergence and diversity-based selection metrics are used to enable PCM to find near-native proteins with well-distributed energy values, while a Pareto-dominance-based archive is proposed to save more potential conformations that can guide the search to more promising conformation areas. The experimental results on thirty-four benchmark proteins demonstrate the significant superiority of PCM in comparison with other single, multiple, and many-objective evolutionary algorithms. Additionally, the inherent characteristics of iterative search of PCM can also give more insights into the dynamic progress of protein folding besides the final predicted static tertiary structure. All these confirm that PCM is a fast, easy-to-use, and fruitful solution generation method for PSP.

Protein structure prediction (PSP) problems are a major biocomputing challenge, owing to its scientific intrinsic that assists researchers to understand the relationship between amino acid sequences and protein structures, and to study the function of proteins. Although computational resources increased substantially over the last decade, a complete solution to PSP problems by computational methods has not yet been obtained. Using only one energy function is insufficient to characterize proteins because of their complexity. Diverse protein energy functions and evolutionary computation algorithms have been extensively studied to assist in the prediction of protein structures in different ways. Such algorithms are able to provide a better protein with less computational resources requirement than deep learning methods. For the first time, this study proposes a many-objective PSP (MaOPSP) problem with four types of objectives to alleviate the impact of imprecise energy functions for predicting protein structures. A many-objective evolutionary algorithm (MaOEA) is utilized to solve MaOPSP. The proposed method is compared with existing methods by examining 34 proteins. An analysis of the objectives demonstrates that our generated conformations are more reasonable than those generated by single/multiobjective optimization methods. Experimental results indicate that solving a PSP problem as an MaOPSP problem with four objectives yields better PSPs, in terms of both accuracy and efficiency. The source code of the proposed method can be found athttps://toyamaailab.github.io/sourcedata.html.