Inbarasan Muniraj

Photoacoustic microscopy (PAM) is often implemented with a reduced number of sensors to minimize system complexity and cost. However, this results in sparse data acquisition and limited-view detection, which introduces severe artifacts and degrades image quality, consequently hindering the accurate visualization of fine vascular structures. In this study, we propose to use a hybrid image reconstruction framework that combines a nonlinear Lucy-Richardson deconvolution algorithm with an adaptive Kalman filter to enhance vascular detail under limited-view conditions. The proposed method is evaluated using an optical-resolution PAM dataset, demonstrating improved reconstruction fidelity and diagnostic accuracy for both preclinical and clinical imaging applications. Additionally, we benchmark our approach against a recently developed Swin Transformer-based image restoration model to assess its effectiveness in enhancing image clarity and vascular structure delineation. Compared to noisy PAM vessel images, the proposed method achieves a reduction of up to 95% in mean squared error (MSE) and nearly a 97% improvement in structural similarity index measure (SSIM).

Due to its detailed spatial mapping capability, three-dimensional (3D) imaging has became an indispensable tool for numerous scientific applications, including medical diagnostics, industrial inspection, autonomous navigation, environmental monitoring, etc. Conventional 3D imaging techniques often suffer from limitations such as poor resolution,and sensitivity, especially in low-light or photon-starved imaging conditions. A photon-counted imaging (PCI) system is often preferred to overcome this limitation. It is known that PCI systems leverage photon detection techniques to capture high-resolution 3D spatial information with exceptional sensitivity and resolution. To achieve this, single-photon detectors are used to count/detect (typically from single to a few hundred) photons from extremely low light level imaging conditions. This paper provides a comprehensive overview of PCI-based 3D imaging systems aimed at overcoming the limitations of the conventional 3D imaging systems. In addition to this, we also discuss some of the add-on applications of PCI systems, including information security, denoising, and resolution enhancement. This article comprises three major sections with four topics to provide an overview of recent research activities using photon-counted imaging systems.

A transport of intensity equation (TIE) based three-dimensional (3D) fluorescence imaging using photon-counting detection was proposed for low light-level bioimaging applications. The number of photons required to achieve the quantitative phase measurement using the proposed system was investigated experimentally and numerically. Results show that the feasibility of reconstructing phase information with an average number of photons is greater than about 9 per pixel in our plant cells.

Cholangiocarcinoma is one of the rarest yet most aggressive cancers that has a low 5-year survival rate (2% - 24%) and thus often requires an accurate and timely diagnosis. Hyperspectral Imaging (HSI) is a recently developed, promising spectroscopic-based non-invasive bioimaging technique that records a spatial image (x, y) together with wide spectral (λ) information. In this work, for the first time we propose to use a three-dimensional (3D)U-Net architecture for Hyperspectral microscopic imaging-based cholangiocarcinoma detection and classification. In addition to this architecture, we opted for a few preprocessing steps to achieve higher classification accuracy (CA) with minimal computational cost. Our results are compared with several standard unsupervised and supervised learning approaches to prove the efficacy of the proposed network and the preprocessing steps. For instance, we compared our results with state-of-the-art architectures, such as the Important-Aware Network (IANet), the Context Pyramid Fusion Network (CPFNet), and the semantic pixel-wise segmentation network (SegNet). We showed that our proposed architecture achieves an increased CA of 1.29% with the standard preprocessing step i.e., flat-field correction, and of 4.29% with our opted preprocessing steps.

We demonstrate that the combination of Super Linear Iterative Clustering and Earth Mover’s Distance efficiently segments tumours from the MRI dataset. Despite using a smaller training dataset our approach achieves an accuracy of 86.2%.