Engineering approach vs reservoir computing to detect electromagnetic interference noise in photoacoustic imaging

Abstract

Electromagnetic interference noise is common in photoacoustic endoscopy, which needs to be resolved before its use in clinics. Nevertheless, little has been known in the literature about how to handle the noise computationally. In this work, we investigate two approaches to deal with the noise. One is an engineering approach. Based on the image structure, we consider the gradient descent method in the polar coordinates for low-level noise and confirm its computational efficiency and accuracy in comparison with the gradient descent method in the Cartesian coordinates. The other is reservoir computing utilizing self-organizing maps for high-level noise. This approach harnesses the dynamic processing capabilities of recurrent neural networks to manage the complexity and variability of intense noise patterns. We believe this work suggests an effective extension of reservoir computing to photoacoustic imaging.