Automatic Detection of Acute Leukemia (ALL and AML) Utilizing Customized Deep Graph Convolutional Neural Networks

Area of Science:

• Medical Imaging
• Computational Biology
• Machine Learning

Background:

• Leukemia, a blood cancer, poses significant health risks including infections and mortality.
• Early diagnosis of leukemia is crucial for effective treatment and improved patient outcomes.
• Machine learning and deep learning offer promising avenues for enhancing diagnostic accuracy.

Purpose of the Study:

• To develop an automated, end-to-end deep learning model for diagnosing acute lymphocytic leukemia (ALL) and acute myeloid leukemia (AML).
• To evaluate the model's performance, resilience to noise, and potential as a clinical decision support tool.

Main Methods:

• A novel deep model architecture was designed, fusing graph theory with a convolutional neural network (CNN).
• The model incorporates six graph convolutional layers and a Softmax layer for classification.
• A dataset of 670 ALL and AML images from 44 patients was utilized for training and validation.

Main Results:

• The proposed deep model achieved a high classification accuracy of 99% for ALL and AML.
• A kappa coefficient of 0.85 was obtained, indicating strong agreement in classification.
• The model demonstrated significant resilience in noisy conditions, maintaining over 90% accuracy at a 0 dB signal-to-noise ratio (SNR).

Conclusions:

• The developed deep learning model shows exceptional accuracy and robustness in differentiating between ALL and AML.
• Its resilience to noise suggests practical applicability in diverse clinical imaging scenarios.
• The model holds potential as a valuable tool to aid clinicians in the early and accurate diagnosis of acute leukemia subtypes.