An Efficient Detection and Classification of Acute Leukemia Using Transfer Learning and Orthogonal Softmax...

Area of Science:

• Medical Imaging
• Computational Biology
• Machine Learning in Healthcare

Background:

• Microscopic analysis of blood cells is crucial for early diagnosis of hematological disorders like leukemia.
• Traditional deep Convolutional Neural Networks (CNNs) often overfit small medical image datasets, limiting their diagnostic accuracy.
• Existing models struggle with the challenge of classifying Acute Lymphoblastic Leukemia (ALL) and Acute Myelogenous Leukemia (AML) on limited data.

Purpose of the Study:

• To propose a novel and effective deep learning model for the classification and detection of ALL and AML.
• To address the overfitting issue commonly encountered with small medical image datasets.
• To enhance classification performance and computational efficiency in leukemia detection.

Main Methods:

• Development of a novel Orthogonal SoftMax Layer (OSL)-based Acute Leukemia detection model.
• Integration of OSL with ResNet18 for deep feature extraction and classification, enforcing orthogonality in weight vectors.
• Inclusion of additional dropout and ReLU layers to improve network speed and performance.

Main Results:

• The proposed OSL-ResNet18 model demonstrated superior performance in classifying ALL and AML on benchmark datasets (ALLIDB1, ALLIDB2, C_NMC_2019, ASH).
• The model effectively overcomes the overfitting problem associated with small medical image datasets.
• Experimental results show significant improvements over existing comparative models in leukemia detection accuracy and efficiency.

Conclusions:

• The OSL-ResNet18 model offers a robust and efficient solution for early leukemia diagnosis, particularly in resource-limited scenarios with small datasets.
• Orthogonal SoftMax Layer integration significantly enhances feature discrimination and computational efficiency.
• This approach holds promise for improving diagnostic capabilities in hematological disorders through advanced deep learning techniques.