Deep learning acceleration of multiscale superresolution localization photoacoustic imaging

Abstract

A superresolution imaging approach that localizes very small targets, such as red blood cells or droplets of injected photoacoustic dye, has significantly improved spatial resolution in various biological and medical imaging modalities. However, this superior spatial resolution is achieved by sacrificing temporal resolution because many raw image frames, each containing the localization target, must be superimposed to form a sufficiently sampled high-density superresolution image. Here, we demonstrate a computational strategy based on deep neural networks (DNNs) to reconstruct high-density superresolution images from far fewer raw image frames. The localization strategy can be applied for both 3D label-free localization optical-resolution photoacoustic microscopy (OR-PAM) and 2D labeled localization photoacoustic computed tomography (PACT). For the former, the required number of raw volumetric frames is reduced from tens to fewer than ten. For the latter, the required number of raw 2D frames is reduced by 12 fold. Therefore, our proposed method has simultaneously improved temporal (via the DNN) and spatial (via the localization method) resolutions in both label-free microscopy and labeled tomography. Deep-learning powered localization PA imaging can potentially provide a practical tool in preclinical and clinical studies requiring fast temporal and fine spatial resolutions.

Deep learning alignment of bidirectional raster scanning in high speed photoacoustic microscopy

“…The network contained roughly 1.6 million learnable parameters (Fig. 2 b) 37 .

Figure 2 Visual representation of MS-FD-U-Net GAN.

…”

“…To obtain a dataset for training DNNs, our previously reported OR-PAM system with a water-immersible galvanometer (OptichoM, Opticho, South Korea) was used in the current work 37 . It employed a fast nanopulse laser system with a maximum PRR of 600 kHz (VPFL-G-10, Spectra-Physics, USA) to induce PA waves.…”

“…Deep learning (DL), which is a rapidly developing field, has addressed the challenges associated with enhancement, translation, segmentation, classification, and medical decision-making in biomedical imaging tools such as PAI, computed tomography (CT), ultrasound imaging, magnetic resonance angiography (MRI), and optical microscopy 25 – 37 . Specifically, an interesting application of DL techniques lies in improving the imaging system performance, such as quality and speed 26 , 27 , 37 , 38 . Here, we present a scanning framework based on deep neural networks (DNNs) to halve the imaging acquisition time by correcting the misaligned OR-PAM images obtained via bidirectional scanning.…”

Whole-Body Photoacoustic Imaging Techniques for Preclinical Small Animal Studies

“…In addition to traditional delay-and-sum or model-based algorithms, deep learning techniques have been widely investigated in recent years [ 90 ]. Deep learning methods have been variously applied in PAI, including in image reconstruction with improved resolution [ 91 , 92 ] or signal-to-noise ratio [ 93 ], quantitative image acquisition [ 94 ], correction of the speed of sound [ 95 ], and image segmentation [ 96 ]. These enhancements can also be applied to preclinical small animal imaging for better spatiotemporal resolution, which will expand the application area of PAI systems.…”

Fully integrated photoacoustic microscopy and photoplethysmography of human in vivo

“…Thus, the capability of PAI renders it useful for imaging blood vessels and monitoring hemodynamics by highlighting hemoglobin at visible wavelengths [8] . In addition, PAI enables multiscale imaging from microscopy to clinical applications, depending on which optical and ultrasonic subsystems are combined [9] , [10] , [11] , [12] , [13] , [14] , [15] . Most clinical studies have been conducted in the form of photoacoustic tomography with a high intensity pulsed laser and medical ultrasound machines that are widely used in hospitals and clinics [16] , [17] , [18] , [19] .…”