Investigations into hyperspectral and high-resolution imaging of layered biosamples

• Main Author: Antony, Maria Merin
• Other Authors: Murukeshan Vadakke Matham
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2025
• Subjects: Agricultural Sciences; Engineering; Physics; Hyperspectral imaging; High-resolution imaging; Optical imaging; Layered biosamples; Bioimaging
• Online Access: https://hdl.handle.net/10356/182540
• Institution: Nanyang Technological University

The field of diagnostic imaging has experienced a rapid growth in recent years driven by the increased demand for better diagnosis and the desire to understand biological mechanisms at the microscopic level. It is predicted that the future diagnostic imaging techniques would require accurate diagnosis for well-defined predictions and classifications. Among the diagnostic imaging techniques, those using the optical wavelength regime are more preferred because of its non-invasive and non-ionizing properties that can reduce the exposure to harmful radiations. However, there are numerous challenges to overcome before optical imaging techniques can be used effectively in the automated diagnostic assessment of biological tissues, especially in thick and layered tissues. Processes such as high absorption, strong scattering, specular reflections, and autofluorescence makes the optical imaging of such samples tedious. For thick absorptive biological samples, the multiple scattering and absorption events occurring during the photon propagation, limit the penetration depth to the optical diffusion limit and the possibility to perform direct imaging in thicker samples. On the other hand, the high transmission and specular nature associated with transparent biosamples make it difficult for them to be imaged using light in reflection configuration. Thus, the current scenario requires specialised systems and trained personnel to image specific biosamples; absorptive or transparent. The operator skill dependence of the current techniques will increase the chance of diagnostic errors, which affect the accurate diagnosis of the disease state and can be a serious threat to survival. In brief, the currently available technologies for diagnostic imaging of such samples are limited by, (i) low working distance and small field of view for high-resolution measurements (ii) lack of chemical specificity, (iii) longer image acquisition time, (iv) high operator skill dependence leading to lack of automation possibilities, (v) destructive nature (require labelling or clearing), and (vi) lower imaging depth due to high absorption and scattering. Hence, there is a genuine need for novel optical imaging concepts and systems to be developed for the high-resolution imaging of both absorptive and transparent multilayered biosamples that can subdue these limitations. This research explores theoretical and experimental investigations on the various processes to mitigate the current limitations as mentioned earlier while imaging such biosamples. Based on the conducted literature survey, two multi-layered biosamples, namely the cornea of the eye (optically transparent) and plant leaf (optically absorptive), have been identified, whose optical imaging still faces challenges discussed earlier. The systems developed in this research are focused on overcoming the challenges associated with the diagnostic imaging of these samples. In this context, one of the major focuses of this research is to investigate the potential of two primary imaging modalities: spectral imaging and high-resolution imaging using structured illumination embedded with speckles, individually or by using a bimodal approach for sample characterisation and related diagnosis. Previous research has shown that the unique spectral signatures associated with the biosample can play a significant role in the accurate assessment of the underlying conditions of the sample. Thus, the inclusion of spectral information in high-resolution imaging techniques can ultimately lead to accurate diagnosis in layered biosamples. The realisation of such experimental systems that records the spectral characteristics along with high-resolution spatial information can prepare diagnostic imaging tools for facing the future digital world. In this context, the first objective of this research is to develop a non-contact, large-area surface hyperspectral imaging technique offering high spectral resolution. The research aims at designing the optics (including lens parameters and illumination schemes for large-area monitoring) and developing the data processing algorithms. An automated spectral imaging system to continuously monitor the small changes in the biosamples was developed. The developed configuration can acquire hyperspectral images with a spatial resolution of ~140 µm at a working distance of ~1 m and spectral resolution of ~1.4 nm. Also, the system can process the data at high speeds using the spectral index approach (7 GB size data was processed in less than a minute on a i7 processor running at 1.9 GHz speed with 32 GB RAM) and is therefore envisaged to reduce the processing time in monitoring large areas (such as vertical hydroponic farms spanning an area of the order of ~15 m2 - 20 m2 in a single scan). These systems also helped in reducing the dependence on human labour and skills in decision making. Hence, this system can also help in the automated and accurate characterisation or disease diagnosis in the samples using appropriate spectral libraries developed during the research. Investigating the processes and mechanisms occurring in a stressed and healthy biosample is crucial. This leads to the second main objective of this research as the development of a high-resolution imaging system to image multilayered biosamples with high lateral and axial resolution with long working distances enabling non-contact and non-invasive imaging. The light propagation in layered biological tissues is investigated using Monte Carlo simulations considering the specific cases for both optically transparent and absorptive regions using unstructured (conventional) and structured laser light. The simulation results illustrated that the use of structured light for imaging such thick multilayered samples proved to be better for diagnostic imaging. This approach was followed for the realization of an imaging system employing 400 structured illumination patterns embedded with speckles (named as embedded speckle structured illumination microscope or in short as ES-SIM). The conceptualized and developed imaging configuration can image the biosamples with a lateral resolution of 1 µm (in the present case, this was limited by pixel size of the camera used) and an axial resolution (Δz) of ~3 µm, at long working distances (greater than 1 cm) and field of view (FOV) of 0.5 mm × 0.5 mm using a 20× objective lens (0.45 NA, 19 mm WD). High axial and lateral resolutions offered by this technique allow for accurate 3D reconstruction enabling the understanding of various morphological changes that can occur in the biosample as the disease progresses. The depth of imaging achieved for absorptive sample is ~ 80 µm and for transparent sample is ~ 900 µm. The capability of the developed high-resolution imaging configuration for both absorptive and transparent biosamples with multiple layers for diagnostic applications such as necrosis detection and corneal characterisation were demonstrated with test samples. Further to optimise the image acquisition time and reduce imaging artefacts, a computational imaging algorithm was developed (termed as FAST ES-SIM) which reduced the image acquisition time by 10 times and improved the speckle contrast by ~5 times. The final objective of this thesis is to conceptualise and develop a bimodal imaging system integrating the spectroscopic and high-resolution imaging of layered biosamples for diagnostic applications along the wavelength range from 400 nm to 1000 nm. Two microscopic configurations namely, structured illumination based hyperspectral microscope (SIHM) and speckle based hyperspectral microscope (SHSM) integrating the high resolution imaging and spectral imaging aspects were conceptualised and fabricated. The developed SIHM configuration offers a spatial resolution of ~586 nm (lateral) and ~5 µm (axial) at the wavelength of 640 nm using an objective lens (50×, 0.55 NA, 13 mm WD, Apochromat). Structured illumination patterns enhanced the lateral resolution of the system beyond the theoretical diffraction limit of 710 nm (at 0.55 NA and λ = 640 nm) and thereby improving the effective NA of the system by 1.2 times. The system also offers a high spectral resolution of ~1.4 nm. As an application, the structural and spectral changes occurring during the necrosis and chlorosis processes in a leaf sample was investigated using SIHM configuration. The developd SIHM used pushbroom configuration for spectral scanning which demanded more acquisition time for each pattern projection (for 9 structured illumination patterns, approx. 4 min). To reduce the acquisition time, a single scan hyperspectral microscope is developed termed as SHSM using the wavelength dependent nature of laser speckle generated using a static diffuser. This configuration reduced the image acquisition time by ~4.5 times. The identified research problems that are targeted to be improved during this research included: Non-invasive corneal layer characterisation for corneal quality assessment of the eye and crop monitoring with associated disease diagnosis in hydroponics, specifically focusing on diseases such as necrosis and chlorosis. It is envisioned that the contributions from this research can help in improving diagnostic imaging of multi-layered biosamples, especially for applications in (i) biomedical area such as corneal characterisation and disease detection, and (ii) for comprehensive in-situ crop monitoring of plants. The developed systems are anticipated to drive a paradigm shift not only in the biomedical and agricultural industries but also in other sectors such as automotive, marine, and semiconductor industries in the future.