Machine Learning for Diagnosis and Screening of Chronic Lymphocytic Leukemia Using Routine Complete Blood Count (CBC)...

Area of Science:

• Hematology
• Medical Informatics
• Computational Biology

Background:

• Chronic lymphocytic leukemia (CLL) is the most common leukemia subtype, accounting for 25-30% of all leukemia cases.
• Despite its prevalence, there is a significant gap in the application of artificial intelligence (AI) for CLL diagnosis.
• Existing diagnostic methods may be resource-intensive or delayed, impacting patient outcomes.

Purpose of the Study:

• To investigate the efficacy of data-driven AI techniques for diagnosing CLL using only complete blood count (CBC) parameters.
• To explore the potential of leveraging immune dysfunctions reflected in routine CBC data for AI-based CLL detection.
• To develop robust and accurate AI classifiers for early CLL diagnosis.

Main Methods:

• Utilized statistical inferences and four distinct feature selection methods.
• Employed multistage hyperparameter tuning to optimize classifier performance.
• Developed and evaluated AI models including Quadratic Discriminant Analysis (QDA), Logistic Regression (LR), and XGboost (XGb).

Main Results:

• Achieved high classification accuracies: 97.05% for QDA, 97.63% for LR, and 98.62% for XGboost.
• Demonstrated that AI models analyzing CBC data can effectively identify CLL.
• Highlighted the potential of these methods in reflecting underlying immune dysfunctions.

Conclusions:

• AI-driven analysis of CBC data presents a promising avenue for the timely diagnosis of CLL.
• These methods offer a cost-effective and resource-efficient approach to cancer detection.
• Improved diagnostic capabilities can lead to better patient care and outcomes.