Dynamic modeling of photoacoustic sensor data to classify human blood samples

Area of Science:

• Biomedical Optics
• Signal Processing
• Hematology

Background:

• Photoacoustic (PA) imaging offers diagnostic potential in biomedicine.
• Automated analysis of PA signals for qualitative results remains challenging.
• Accurate classification of blood physiological status is crucial for disease diagnosis.

Purpose of the Study:

• To develop a dynamic modeling scheme for photoacoustic sensor data.
• To classify whole human blood samples based on physiological status.
• To evaluate the performance of the proposed method against conventional techniques.

Main Methods:

• A state-space model was estimated using a subspace method for 35 whole human blood samples.
• Blood samples were classified using model parameters and linear discriminant analysis.
• Performance was compared with time-domain, frequency-domain features, and an autoregressive-moving-average model.

Main Results:

• The proposed analysis successfully predicted five blood classes: healthy (women and men), microcytic anemia, macrocytic anemia, and leukemia.
• The dynamic modeling scheme significantly outperformed conventional signal processing methods.
• The method demonstrated high accuracy in classifying blood samples based on hematological status.

Conclusions:

• The dynamic modeling scheme provides a robust and automated approach for analyzing photoacoustic data.
• This method shows promise for point-of-care devices in detecting hematological diseases.
• The findings support the use of photoacoustic signal analysis for advanced medical diagnostics.