Blood cancer diagnosis using ensemble learning based on a random subspace method in laser-induced breakdown...

Area of Science:

• Analytical Chemistry
• Biomedical Spectroscopy
• Machine Learning for Diagnostics

Background:

• Accurate diagnosis of blood cancers, distinguishing them from healthy controls and identifying specific types, remains a significant clinical challenge.
• Traditional chemometrics methods combined with LIBS show promise but are limited by spectral data redundancy and noise, impacting diagnostic accuracy.
• Existing methods struggle with the complexity of differentiating various blood cancer subtypes.

Purpose of the Study:

• To develop and validate an enhanced diagnostic approach for blood cancers using LIBS coupled with ensemble learning.
• To improve the accuracy of discriminating blood cancers from healthy samples.
• To enhance the capability of identifying specific blood cancer types, including leukemia, multiple myeloma, and lymphoma.

Main Methods:

• Laser-Induced Breakdown Spectroscopy (LIBS) was employed for serum sample analysis.
• Serum samples were prepared on a boric acid substrate for spectral data acquisition.
• An ensemble learning approach, specifically the Random Subspace Method (RSM) combined with Linear Discriminant Analysis (LDA) (RSM-LDA), was utilized for data analysis.

Main Results:

• The RSM-LDA model achieved an average accuracy of 98.34% for blood cancer detection, a significant improvement over LDA alone (94.45%).
• Variable importance analysis identified key spectral lines (Na, K, Mg, Ca, H, O, N, C-N) crucial for cancer recognition.
• The RSM-LDA model improved blood cancer type identification accuracy from 80.4% to 91.0% compared to LDA alone.

Conclusions:

• LIBS combined with the RSM-LDA model offers a robust and accurate method for blood cancer diagnosis.
• This approach effectively differentiates blood cancers from healthy controls and accurately identifies specific cancer subtypes.
• The findings suggest a promising new tool for clinical hematology, enhancing diagnostic precision and efficiency.