Identification of Leukemia Subtypes from Microscopic Images Using Convolutional Neural Network

Area of Science:

• Medical Imaging
• Computational Biology
• Oncology

Background:

• Leukemia is a fatal cancer with four main subtypes: acute lymphoid, acute myeloid, chronic lymphoid, and chronic myeloid.
• Accurate and early diagnosis of leukemia subtypes is crucial for effective treatment and patient outcomes.
• Microscopic blood cell images offer a potential source for automated leukemia diagnosis.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for the automated diagnosis of all four leukemia subtypes using microscopic blood cell images.
• To investigate the impact of data augmentation techniques on improving the performance of the diagnostic model.
• To compare the performance of the proposed convolutional neural network (CNN) model against traditional machine learning algorithms.

Main Methods:

• Utilized two public leukemia datasets (ALL-IDB and ASH Image Bank) comprising microscopic blood cell images.
• Applied seven distinct image transformation techniques for data augmentation to synthetically increase the training dataset size.
• Designed and implemented a custom CNN architecture for leukemia subtype classification and compared it with Naive Bayes, SVM, k-NN, and Decision Tree algorithms.
• Employed 5-fold cross-validation for rigorous model performance evaluation.

Main Results:

• The CNN model achieved 88.25% accuracy in distinguishing leukemia from healthy samples.
• The CNN model demonstrated 81.74% accuracy in multiclass classification of all four leukemia subtypes.
• The proposed CNN model outperformed other traditional machine learning algorithms in leukemia diagnosis tasks.

Conclusions:

• Convolutional neural networks, enhanced by data augmentation, provide a robust and accurate method for diagnosing all leukemia subtypes from microscopic blood cell images.
• The developed CNN model shows significant potential for improving automated leukemia diagnostic systems.
• This approach offers a promising tool for early and precise detection of leukemia, aiding clinical decision-making.