Advances in Photoacoustic Imaging Reconstruction and Quantitative Analysis for Biomedical Applications

Photoacoustic imaging (PAI) represents an innovative biomedical imaging modality that harnesses the advantages of optical resolution and acoustic penetration depth while ensuring enhanced safety. Despite its promising potential across a diverse array of preclinical and clinical applications, the clinical implementation of PAI faces significant challenges, including the trade-off between penetration depth and spatial resolution, as well as the demand for faster imaging speeds. This paper explores the fundamental principles underlying PAI, with a particular emphasis on three primary implementations: photoacoustic computed tomography (PACT), photoacoustic microscopy (PAM), and photoacoustic endoscopy (PAE). We undertake a critical assessment of their respective strengths and practical limitations. Furthermore, recent developments in utilizing conventional or deep learning (DL) methodologies for image reconstruction and artefact mitigation across PACT, PAM, and PAE are outlined, demonstrating considerable potential to enhance image quality and accelerate imaging processes. Furthermore, this paper examines the recent developments in quantitative analysis within PAI, including the quantification of haemoglobin concentration, oxygen saturation, and other physiological parameters within tissues. Finally, our discussion encompasses current trends and future directions in PAI research while emphasizing the transformative impact of deep learning on advancing PAI.

💡 Research Summary

Photoacoustic imaging (PAI) is emerging as a hybrid biomedical modality that uniquely combines the high optical contrast of light absorption with the deep penetration and spatial resolution of ultrasound. This review provides a comprehensive overview of the fundamental physics, major system implementations, reconstruction strategies, quantitative analysis techniques, and future directions that together shape the current state and clinical translation prospects of PAI.

The authors first revisit the photoacoustic effect, emphasizing the dual requirements of stress confinement (laser pulse shorter than the acoustic relaxation time) and thermal confinement (negligible heat diffusion during the pulse). Under these conditions the initial pressure rise p₀ is linearly proportional to the absorbed optical energy density via the dimensionless Grüneisen parameter Γ, establishing the theoretical basis for quantitative imaging.

Three primary PAI architectures are examined in detail: photoacoustic computed tomography (PACT), photoacoustic microscopy (PAM), and photoacoustic endoscopy (PAE). PACT employs a wide‑field, unfocused laser illumination and a large‑aperture ultrasonic array (linear, hemispherical, or ring‑shaped) to acquire volumetric data from centimeters deep within tissue. It offers moderate spatial resolution (50–200 µm) and real‑time imaging speeds, making it suitable for whole‑organ applications such as breast cancer screening or brain functional mapping. However, limited angular coverage and finite detector bandwidth introduce reconstruction artefacts and constrain resolution.

PAM is divided into optical‑resolution (OR‑PAM) and acoustic‑resolution (AR‑PAM) variants. OR‑PAM achieves sub‑5 µm lateral resolution by tightly focusing the excitation beam, but its imaging depth is limited to 1–2 mm due to optical scattering. AR‑PAM relaxes the optical focus, allowing acoustic focusing to dictate resolution (≈50 µm) and extending depth to 3–10 mm, at the cost of higher background signals. PAM excels in microvascular, perfusion, and oxygenation studies of superficial tissues.

PAE integrates a miniature optical fiber and an ultrasound transducer into an endoscopic probe, enabling intravascular or intracavitary imaging with depth up to several centimeters. While promising for real‑time plaque characterization and guided interventions, challenges remain in miniaturizing high‑frequency transducers and maintaining sufficient optical fluence within the confined geometry.

Reconstruction methods are a central theme. Classical algorithms—time‑reversal, filtered back‑projection, model‑based iterative reconstruction (e.g., total variation regularization)—are described, together with their limitations: ill‑posedness, sensitivity to noise, and computational burden that hinder real‑time deployment. The review then surveys the rapid rise of deep learning (DL) approaches. Convolutional encoder‑decoder networks (U‑Net), generative adversarial networks (GAN), and physics‑informed neural networks have been applied to (i) artefact suppression, (ii) accelerated reconstruction from sparsely sampled data, and (iii) joint reconstruction‑quantification pipelines. By embedding physical constraints (e.g., wave equation loss, non‑negativity) into the loss function, DL models achieve higher PSNR/SSIM than conventional methods while delivering frame rates exceeding 30 fps, a critical step toward clinical workflow integration.

Quantitative PAI is discussed with emphasis on multispectral unmixing of endogenous chromophores (oxy‑ and deoxy‑hemoglobin, melanin, lipids) and exogenous agents. The linear relationship between absorbed energy and pressure enables extraction of absolute chromophore concentrations and blood oxygen saturation (sO₂) when multiple wavelengths are employed. The authors highlight challenges such as fluence heterogeneity, spectral coloring, and non‑linear thermal effects that introduce bias. Recent works combine DL‑based fluence correction with model‑based spectral inversion, achieving concentration errors below 5 % in vivo. Extensions to quantify total hemoglobin, metabolic rate of oxygen, and temperature are also presented.

Contrast agent development is reviewed, ranging from clinically approved indocyanine green (ICG) to gold nanorods/nanocages, semiconducting polymer nanoparticles, and perovskite nanocrystals. While metallic nanostructures provide strong, tunable near‑infrared absorption, concerns about long‑term biodistribution and toxicity limit their clinical use. Polymer‑based agents offer better biodegradability and photostability, whereas perovskites deliver exceptionally high absorption coefficients but suffer from lead toxicity and colloidal instability. The authors stress the need for rigorous biocompatibility assessments and regulatory pathways for any new agent.

Finally, the review outlines future research directions essential for clinical translation: (1) hardware integration of ultra‑broadband, high‑frequency ultrasound arrays with high‑pulse‑energy lasers to push resolution and depth simultaneously; (2) creation of large, annotated clinical PAI datasets to train robust, generalizable DL models; (3) design of safe, biodegradable nanocontrast agents with strong NIR absorption and rapid clearance; (4) development of closed‑loop theranostic platforms where quantitative PAI feedback guides photothermal or photodynamic therapy in real time. The authors conclude that deep learning is poised to be the transformative catalyst that bridges the gap between laboratory prototypes and routine clinical practice, by delivering fast, high‑quality reconstructions and reliable quantitative metrics.