An attention-based deep learning for acute lymphoblastic leukemia classification

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Июль 29, 2024

Abstract The bone marrow overproduces immature cells in the malignancy known as Acute Lymphoblastic Leukemia (ALL). In United States, about 6500 occurrences of ALL are diagnosed each year both children and adults, comprising nearly 25% pediatric cancer cases. Recently, many computer-assisted diagnosis (CAD) systems have been proposed to aid hematologists reducing workload, providing correct results, managing enormous volumes data. Traditional CAD rely on hematologists’ expertise, specialized features, subject knowledge. Utilizing early detection can radiologists doctors making medical decisions. this study, Deep Dilated Residual Convolutional Neural Network (DDRNet) is presented for classification blood cell images, focusing eosinophils, lymphocytes, monocytes, neutrophils. To tackle challenges like vanishing gradients enhance feature extraction, model incorporates Blocks (DRDB) faster convergence. Conventional residual blocks strategically placed between layers preserve original information extract general maps. Global Local Feature Enhancement (GLFEB) balance weak contributions from shallow improved normalization. global initial convolution layer, when combined with GLFEB-processed reinforces representations. Tanh function introduces non-linearity. A Channel Spatial Attention Block (CSAB) integrated into neural network emphasize or minimize specific channels, while fully connected transform use a sigmoid activation concentrates relevant features multiclass lymphoblastic leukemia was analyzed Kaggle dataset (16,249 images) categorized four classes, training testing ratio 80:20. Experimental results showed that DRDB, GLFEB CSAB blocks’ discrimination ability boosted DDRNet F1 score 0.96 minimal computational complexity optimum accuracy 99.86% 91.98% stands out existing methods due its high 91.98%, 0.96, complexity, enhanced ability. strategic combination these (DRDB, GLFEB, CSAB) designed address process, leading crucial accurate multi-class image identification. Their effective integration within contributes superior performance DDRNet.

Язык: Английский

---

CoTCoNet: An optimized coupled transformer-convolutional network with an adaptive graph reconstruction for leukemia detection

Computers in Biology and Medicine, Год журнала: 2024, Номер 179, С. 108821 - 108821

Опубликована: Июль 6, 2024

Язык: Английский

---

Pyranet: a novel architectural approach to reduce the effect of unbalanced classes and analysis on leukemia dataset

Soft Computing, Год журнала: 2025, Номер unknown

Опубликована: Май 13, 2025

Язык: Английский

---

TW-YOLO: An Innovative Blood Cell Detection Model Based on Multi-Scale Feature Fusion

Sensors, Год журнала: 2024, Номер 24(19), С. 6168 - 6168

Опубликована: Сен. 24, 2024

As deep learning technology has progressed, automated medical image analysis is becoming ever more crucial in clinical diagnosis. However, due to the diversity and complexity of blood cell images, traditional models still exhibit deficiencies detection. To address detection, we developed TW-YOLO approach, leveraging multi-scale feature fusion techniques. Firstly, CNN (Convolutional Neural Network) convolution poor recognition capabilities for certain features, so RFAConv (Receptive Field Attention Convolution) module was incorporated into backbone model enhance its capacity extract geometric characteristics from cells. At same time, utilizing pyramid architecture YOLO (You Only Look Once), enhanced features at different scales by incorporating CBAM Block Module) detection head EMA (Efficient Multi-Scale Attention) neck, thereby improving ability Additionally, meet specific needs designed PGI-Ghost (Programmable Gradient Information-Ghost) strategy finely describe gradient flow throughout process extracting further model’s effectiveness. Experiments on datasets such as BloodCell-Detection-Dataset (BCD) reveal that outperforms other 2%, demonstrating excellent performance task In addition advancing research, this work offers strong technical support future diagnostics.

Язык: Английский

---

BSNEU-net: Block Feature Map Distortion and Switchable Normalization-Based Enhanced Union-net for Acute Leukemia Detection on Heterogeneous Dataset

Deleted Journal, Год журнала: 2024, Номер unknown

Опубликована: Сен. 25, 2024

Acute leukemia is characterized by the swift proliferation of immature white blood cells (WBC) in and bone marrow. It categorized into acute lymphoblastic (ALL) myeloid (AML), depending on whether cell-line origin lymphoid or myeloid, respectively. Deep learning (DL) artificial intelligence (AI) are revolutionizing medical sciences assisting clinicians with rapid illness identification, reducing workload, enhancing diagnostic accuracy. This paper proposes a DL-based novel BSNEU-net framework to detect leukemia. comprises 4 Union Blocks (UB) incorporates block feature map distortion (BFMD) switchable normalization (SN) each UB. The UB employs union convolution extract more discriminant features. BFMD adapted acquire generalized patterns minimize overfitting, whereas SN layers appended improve model's convergence generalization capabilities. uniform utilization batch across sensitive mini-batch dimension changes, which effectively remedied incorporating an layer. Here, new dataset comprising 2400 smear images ALL, AML, healthy cases proposed, as DL methodologies necessitate sizeable well-annotated combat overfitting issues. Further, heterogeneous 2700 created combining four publicly accessible benchmark datasets cases. model achieved excellent performance 99.37% accuracy 99.44% dataset. comparative analysis signifies superiority proposed methodology comparing schemes.

Язык: Английский

---

Unsupervised attention-guided domain adaptation model for Acute Lymphocytic Leukemia (ALL) diagnosis

Biomedical Signal Processing and Control, Год журнала: 2024, Номер 101, С. 107159 - 107159

Опубликована: Ноя. 19, 2024

Язык: Английский

---