Scholarlink

<p>The rising global frequency of these diseases has contributed to a growing demand for efficient and reliable diagnostic methodologies. Blood specimens have been integral in modern diagnostics and provide a breadth of data regarding potential disease scenarios. Nonetheless, traditional diagnostic technologies and methodologies that rely on human analysis, basic statistical modelling, and expert opinion are inadequate to address the complexity and high-dimensionality of modern biomedical data. Traditional analyses, including linear regression and logistic regression, are interpretable methods but have also struggled with the challenges of feature selection, have the potential of overfitting to the data, and have low suitability for modelling the nonlinear relationship common in blood biomarker data. Machine learning (ML) methods, including Support Vector Machines (SVM), Random Forest (RF), still face difficulties in feature redundancy, interpretability, and computational scaling. Deep learning (DL) models, though powerful, are highly demanding of labelled data and typically opaque, which hinders their adoption in the clinic. Current approaches are often lacking in handling high-dimensional, multivariate data and tend to require manual curation or extensive computational power. In addressing these shortcomings, this work presents an innovative framework of PCA-driven feature optimisation integrated with a Multiscale Echo Point Multilayer Perceptron (MLP) model. This architecture combines convolutional layers in an MLP to detect local and global patterns in the data of a blood sample, increasing pattern recognition and predictive power. The model was tested on a balanced Kaggle dataset consisting of varied biochemical parameters of various blood diseases. Significant contributions are formulating a PCA-enhanced Multiscale neural model, incorporating convolutional learning for enhanced pattern recognition, and exhibiting high classification accuracy and dependability over a range of disease types, which underpins the wider use of ML in biomedical diagnosis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Novel biomarker identification for blood diseases using PCA-enhanced Multiscale neural framework

  • 期刊论文

  • 发表日期：

    2026年4月11日

查看全文
  • Harishna Selvakumar,
  • Gracia Nirmala Rani Duraisamy

摘要

The rising global frequency of these diseases has contributed to a growing demand for efficient and reliable diagnostic methodologies. Blood specimens have been integral in modern diagnostics and provide a breadth of data regarding potential disease scenarios. Nonetheless, traditional diagnostic technologies and methodologies that rely on human analysis, basic statistical modelling, and expert opinion are inadequate to address the complexity and high-dimensionality of modern biomedical data. Traditional analyses, including linear regression and logistic regression, are interpretable methods but have also struggled with the challenges of feature selection, have the potential of overfitting to the data, and have low suitability for modelling the nonlinear relationship common in blood biomarker data. Machine learning (ML) methods, including Support Vector Machines (SVM), Random Forest (RF), still face difficulties in feature redundancy, interpretability, and computational scaling. Deep learning (DL) models, though powerful, are highly demanding of labelled data and typically opaque, which hinders their adoption in the clinic. Current approaches are often lacking in handling high-dimensional, multivariate data and tend to require manual curation or extensive computational power. In addressing these shortcomings, this work presents an innovative framework of PCA-driven feature optimisation integrated with a Multiscale Echo Point Multilayer Perceptron (MLP) model. This architecture combines convolutional layers in an MLP to detect local and global patterns in the data of a blood sample, increasing pattern recognition and predictive power. The model was tested on a balanced Kaggle dataset consisting of varied biochemical parameters of various blood diseases. Significant contributions are formulating a PCA-enhanced Multiscale neural model, incorporating convolutional learning for enhanced pattern recognition, and exhibiting high classification accuracy and dependability over a range of disease types, which underpins the wider use of ML in biomedical diagnosis.