Automated cell-type classification combining dilated convolutional neural networks with label-free acoustic sensing

Area of Science:

• Biomedical Engineering
• Acoustic Sensing
• Artificial Intelligence

Background:

• Label-free acoustic sensing with high-frequency ultrasound can analyze single cells in heterogeneous samples.
• Manual postprocessing of backscattered signals is prone to errors and time-consuming.
• Automated cell classification is crucial for efficient biological analysis.

Purpose of the Study:

• To develop an automated cell-type classification system using label-free acoustic sensing and deep learning.
• To improve the accuracy and efficiency of cell classification by minimizing manual intervention.
• To enable precise identification of cell physical properties for applications like personalized cancer medicine.

Main Methods:

• Utilized a 1D convolutional autoencoder for signal denoising and Gaussian noise injection for data augmentation.
• Employed Convolutional Neural Network (CNN) models for classifying denoised backscattered signals.
• Analyzed signal data using 1D CNNs (waveform, frequency spectrum) and 2D CNNs (spectrogram).

Main Results:

• The proposed system achieved reliable and precise classification of different cell types (RBC, PNT1A) and microspheres.
• Demonstrated the system's robustness to noise through data augmentation.
• Evaluated the effectiveness of CNN models and data representations against baseline methods.

Conclusions:

• The developed automated system accurately classifies cell types based on acoustic backscattered signal patterns.
• This label-free, AI-powered approach offers a robust and efficient method for cell analysis.
• The system holds potential for advancing personalized cancer medicine by identifying cells with distinct physical properties.