Scholarlink

<p>Hematological malignancy classification poses significant challenges in clinical diagnostics, particularly for distinguishing between Chronic Myeloid Leukemia (CML), Chronic Lymphocytic Leukemia (CLL), Acute Myeloblastic Leukemia (AML), and Acute Lymphoblastic Leukemia (ALL). We propose a novel integrated framework combining Fully Convolutional Masked Autoencoder V2 (FCMAE V2) with Windowed Patch Attention Transformer (WPAT) for multi-scale feature extraction, Adaptive Class Distribution Balancing GAN (ACDB-GAN) for addressing dataset imbalance, and Adaptive Local Histogram Enhancement (ALHE) for image preprocessing. Evaluated on the challenging Raabin dataset, our approach achieves 96% classification accuracy, and demonstrates an improvement of approximately 10% over state-of-the-art methods. This synergistic integration of advanced neural architectures with sophisticated data balancing strategies addresses critical limitations in biomedical image analysis, offering enhanced robustness and clinical applicability for leukemia 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

Advanced deep learning framework for automated hematological malignancy classification: integrating FCMAE V2-WPAT with ACDB-GAN for enhanced leukemia subtype detection

  • 期刊论文

  • 发表日期：

    2026年1月16日

查看全文
  • Seyed Vahab Shojaedini,
  • Vahid Reza Nafisi,
  • Manouchehr Eghbal,
  • Mohammad Momenian

摘要

Hematological malignancy classification poses significant challenges in clinical diagnostics, particularly for distinguishing between Chronic Myeloid Leukemia (CML), Chronic Lymphocytic Leukemia (CLL), Acute Myeloblastic Leukemia (AML), and Acute Lymphoblastic Leukemia (ALL). We propose a novel integrated framework combining Fully Convolutional Masked Autoencoder V2 (FCMAE V2) with Windowed Patch Attention Transformer (WPAT) for multi-scale feature extraction, Adaptive Class Distribution Balancing GAN (ACDB-GAN) for addressing dataset imbalance, and Adaptive Local Histogram Enhancement (ALHE) for image preprocessing. Evaluated on the challenging Raabin dataset, our approach achieves 96% classification accuracy, and demonstrates an improvement of approximately 10% over state-of-the-art methods. This synergistic integration of advanced neural architectures with sophisticated data balancing strategies addresses critical limitations in biomedical image analysis, offering enhanced robustness and clinical applicability for leukemia diagnosis.