Medical gene expression feature selection is challenged by the high-dimensional small-sample regime and strong co-expression redundancy, which often yields unstable subsets and brittle trade-offs between predictive performance and compactness. This paper proposes a two-stage framework, Class-guided Joint Hybrid Multi-objective Optimizer (CJHMO), that integrates structural candidate generation with wrapper-based multi-objective optimization. In stage one, a Multi-Scale Co-Expression Attention Network (MSCANet) constructs correlation graphs under multiple thresholds and extracts connected components as co-expression modules. Each module is summarized by its eigengene (the first principal component), and the correlation ratio is used to quantify the association between eigengenes and class labels, producing supervised module scores. These scores are converted into attention weights and propagated to genes for candidate ranking and screening. In stage two, we develop a Dual-Population Heterogeneous Multi-Objective optimizer (DPHMO), where a decomposition-based population emphasizes Pareto-front coverage and global exploration, while an elite-guided particle swarm focuses on local exploitation and refinement. The two populations share an external elite archive (EP) for cross-population information exchange and non-dominated solution maintenance, jointly minimizing classification error rate and feature selection rate. Experiments on multiple public benchmarks with several classifiers demonstrate that CJHMO achieves superior performance-compression trade-offs over representative multi-objective baselines, with improved Pareto quality reflected by HV and IGD.

Deep learning enables the rapid and accurate identification of cell types in peripheral blood smears of acute myeloid leukemia (AML), greatly facilitating its diagnosis. However, current deep learning-based leukocyte classification models face several limitations. Firstly, traditional leukocyte classification models primarily use convolutional neural networks (CNN) as feature extraction modules. While CNN excel at extracting local features of leukocytes, they are less effective at capturing the overall morphological characteristics of the cells. Additionally, pathological leukocyte samples (e.g., certain subtypes in AML) are often scarce, which introduces class bias and significantly reduces the model’s classification performance on these rare categories. To address these challenges, we propose the Triplet Focal Vision Transformer (TFViT) for leukocyte classification in AML within the context of the Internet of Medical Things (IoMT). The TFViT model leverages Vision Transformers (ViT) and self-attention mechanisms to effectively extract high-quality features from leukocyte images, capturing subtle yet critical morphological differences among various subtypes. Moreover, the integration of Triplet Loss enhances the learning of relative relationships between samples, thereby improving the model’s ability to distinguish between similar categories and fine-grained features. Additionally, by incorporating Focal Loss during training, the model prioritizes minority class samples and reduces the dominance of majority classes in the loss calculation. Furthermore, the relative feature constraints of Triplet Loss do not entirely depend on label distribution, providing an alternative approach to mitigating the impact of class imbalance on classification performance. Experimental results demonstrate that the TFViT model achieves outstanding performance in the leukocyte classification task for AML. Specifically, the Precision, Accuracy, and F1-Score values of the TFViT model are 0.957, 0.962, and 0.957, respectively.

Music Emotion Recognition (MER) involves identifying the emotional content conveyed by music. This field is becoming increasingly significant due to its broad range of applications, including music recommendation systems, mood-based playlists, and therapeutic tools. This paper presents a novel MER model designed for song-level analysis, leveraging the Transformer Encoder architecture. The model incorporates various embedding techniques to capture both local and global contexts within musical data, thereby improving the extraction of crucial features for emotion recognition. Additionally, a Self-Attention Pooling Layer is used to effectively integrate and interpret complex musical features. Experiments using the DEAM dataset reveal that this model excels in emotion identification, surpassing existing approaches and offering promising directions for future research in the field of MER.