Deep Learning Based Framework for Detection and Classification of Leukemia Using Microscopic Images

Area of Science:

• Hematology
• Medical Imaging
• Artificial Intelligence

Background:

• Leukemia, a white blood cell cancer, necessitates early diagnosis for effective treatment.
• Microscopy is vital for leukemia diagnosis, but AI can enhance accuracy and efficiency.
• Current automated methods still require medical practitioner oversight.

Purpose of the Study:

• To develop and evaluate a deep learning approach for detecting and classifying leukemia from microscopic images.
• To leverage convolutional neural networks (CNNs) for enhanced leukemia cell identification.
• To assess the model's performance in differentiating between leukemia subtypes.

Main Methods:

• Image pre-processing and data augmentation to expand the microscopic image dataset.
• U-Net model for segmenting leukemia and normal cells for feature extraction.
• CNN-based architecture for classifying leukemia subtypes using the ASH dataset.
• 10-fold cross-validation for performance evaluation.

Main Results:

• Achieved 99.06% accuracy in binary classification and 98.68% in multi-class classification.
• Demonstrated high performance with a recall of 96.74%, precision of 96.83%, and F1-score of 96.77%.
• The model's performance is comparable to existing leukemia detection methods.

Conclusions:

• The proposed framework integrating microscopic imaging and deep learning shows significant potential for computer-aided leukemia diagnosis.
• Further research with larger, diverse datasets and advanced deep learning models is recommended.
• The approach could be extended to diagnose other blood disorders.