[Paper Review] A Diagnostic Model for Acute Lymphoblastic Leukemia Using Metaheuristics and Deep Learning Methods

Rahmani, Amir Masoud, P. Khoshaght+5 more|arXiv (Cornell University)|Jun 2, 2024

Digital Imaging for Blood Diseases4 citations

TL;DR

This study proposes a hybrid diagnostic model for acute lymphoblastic leukemia (ALL) using ResNet-based deep feature extraction, metaheuristic-based feature selection (e.g., binary ant colony optimization), and a multilayer perceptron (MLP) classifier. The model achieves 90.71% accuracy and 95.76% sensitivity on the C-NMC 2019 dataset, outperforming other configurations and demonstrating strong potential for automated ALL diagnosis.

ABSTRACT

Acute lymphoblastic leukemia (ALL) severity is determined by the presence and ratios of blast cells (abnormal white blood cells) in both bone marrow and peripheral blood. Manual diagnosis of this disease is a tedious and time-consuming operation, making it difficult for professionals to accurately examine blast cell characteristics. To address this difficulty, researchers use deep learning and machine learning. In this paper, a ResNet-based feature extractor is utilized to detect ALL, along with a variety of feature selectors and classifiers. To get the best results, a variety of transfer learning models, including the Resnet, VGG, EfficientNet, and DensNet families, are used as deep feature extractors. Following extraction, different feature selectors are used, including Genetic algorithm, PCA, ANOVA, Random Forest, Univariate, Mutual information, Lasso, XGB, Variance, and Binary ant colony. After feature qualification, a variety of classifiers are used, with MLP outperforming the others. The recommended technique is used to categorize ALL and HEM in the selected dataset which is C-NMC 2019. This technique got an impressive 90.71% accuracy and 95.76% sensitivity for the relevant classifications, and its metrics on this dataset outperformed others.

Motivation & Objective

• To address the time-consuming and error-prone nature of manual ALL diagnosis based on blast cell analysis.
• To develop an automated diagnostic system that improves accuracy and sensitivity in detecting acute lymphoblastic leukemia from peripheral blood and bone marrow images.
• To integrate metaheuristic optimization with deep learning for enhanced feature selection and classification performance.
• To evaluate and compare multiple transfer learning models and feature selection techniques in the context of ALL classification.

Proposed method

• A ResNet-based deep neural network is used as a feature extractor to learn high-level representations from ALL and healthy blood cell images.
• Multiple transfer learning models—ResNet, VGG, EfficientNet, and DenseNet—are evaluated for their feature extraction performance.
• A comprehensive set of 10 feature selection methods, including Genetic Algorithm, PCA, ANOVA, Random Forest, Mutual Information, Lasso, XGBoost, Variance, and Binary Ant Colony Optimization, are applied to reduce dimensionality and select relevant features.
• A multilayer perceptron (MLP) classifier is selected as the best-performing model among various classifiers tested.
• The model is trained and evaluated on the C-NMC 2019 dataset, a benchmark dataset for blood cell image classification.
• Hyperparameter tuning and model validation are performed using standard cross-validation and performance metrics such as accuracy and sensitivity.

Experimental results

Research questions

• RQ1 Can a hybrid deep learning and metaheuristic approach improve the accuracy and sensitivity of ALL diagnosis compared to conventional methods?
• RQ2 Which combination of deep feature extractors and feature selection techniques yields the optimal performance in classifying ALL and healthy (HEM) blood cells?
• RQ3 How does the performance of the MLP classifier compare to other classifiers when used with metaheuristic-optimized features?
• RQ4 To what extent do different metaheuristic algorithms outperform traditional feature selection methods in this diagnostic context?
• RQ5 Does the integration of transfer learning with metaheuristic optimization lead to better generalization on the C-NMC 2019 benchmark dataset?

Key findings

• The proposed model achieved a classification accuracy of 90.71% on the C-NMC 2019 dataset, demonstrating high diagnostic reliability.
• The model recorded a sensitivity of 95.76%, indicating strong detection capability for blast cells in ALL cases.
• The multilayer perceptron (MLP) classifier outperformed all other classifiers in terms of both accuracy and sensitivity.
• The binary ant colony optimization algorithm was among the most effective metaheuristic feature selection methods tested.
• ResNet-based feature extraction provided superior performance compared to VGG, EfficientNet, and DenseNet in this specific diagnostic task.
• The integration of metaheuristic feature selection with deep learning significantly improved model performance over baseline configurations.