Fourier transform infrared spectra and machine learning to detect low HER2 expression in breast cancer plasma.

BACKGROUND

In patients with breast cancer, tumors showing low human epidermal growth factor receptor 2 (HER2) expression may not demonstrate clinical benefits from chemotherapy. Since traditional diagnostic methods for detecting HER2 expression require invasive tissue biopsies, we propose a less invasive approach that combines attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy with machine learning to detect breast cancer with low plasma HER2 expression.

METHODS

The leftover heparinized plasma with low HER2 expression from 55 breast cancer patients and 32 healthy controls were performed with ATR-FTIR spectrometer. The ten protocol was applied to preprocessed data analysis. Then, machine learning models such as partial least squares-discriminant analysis (PLS-DA) and neural network were performed. The analytical performance was calculated for accuracy, sensitivity and specificity of the predicted model of detection.

RESULTS

The infrared spectra of low HER2 expression from 55 breast cancer samples and 32 control samples were obtained and analyzed in the 1400 - 1000 cm-1, which is related to the HER2 extracellular domain structure. The neural network models achieved higher discriminative accuracy, sensitivity, and specificity at 78%, while PLS-DA showed 65% accuracy, 71% sensitivity, and 56% specificity.

CONCLUSIONS

This approach has the potential to detect low HER2 expression in less invasive samples. However, validation through larger-scale clinical trials should be considered to achieve more efficiency.