Simultaneous label-free autofluorescence multiharmonic (SLAM) microscopy produces multimodal high-resolution images with rich functional and structural information from living tissue. Here we present a deep learning framework for human breast cancer-related optical biomarker discovery based on SLAM. This framework consists of three stages: self-supervised consistency training for image representation learning at multiple scales; cancer region identification by Multiple Instance Learning; optical biomarker discovery based on channel-wise attribution maps. This study demonstrates the capability of AI-assisted SLAM microscopy in capturing rich information from living tissue and extracting relationships between optical features with human breast cancer, which can be extended to various types of diseases and treatment conditions.

The most prevalent ovarian cancers, high-grade serous carcinomas (HGSCs), begin as lesions in the fallopian tubes. There is a need for tools that examine the fallopian tubes for early-stage ovarian cancers. We hypothesize an endoscopic imaging approach combining structural and functional imaging techniques (optical coherence tomography (OCT) and autofluorescence imaging (AFI) respectively) will be able to visualize cancerous or pre-cancerous changes in the fallopian tubes for early ovarian cancer detection. We present findings from an on-going ex vivo imaging study of fallopian tubes including sample cases demonstrating various cancers and preliminary features of interest.

Recent research has confirmed that rectal cancer patients achieving complete tumor destruction to pre-surgery therapy can safely avoid surgery altogether. Unfortunately, this promising treatment option remains limited due to the inability of existing imaging techniques to identify patients with pathological complete response (pCR). We developed a new co-registered acoustic-resolution photoacoustic microscopy (AR-PAM) and ultrasound system to image rectal tumors undergoing treatment. We then built a custom pattern-recognition convolution neural network (CNN) to provide quantitative image interpretation of AR-PAM. Initial testing of the imaging system and neural network (AR-PAM-CNN) demonstrated promising findings, specifically that tumor beds revert from oncologically-disrupted microvascular patterns to highly regular architecture typical of the normal rectum in the setting of pCR to therapy.

We present a simulation study on the design of a multi-modal high-density hybrid diffuse optical tomographic probe for monitoring infants with congenital heart disease (CHD) before, during, and after surgery. Different probe designs are evaluated based on the signal distribution and sensitivity profile generated using a Monte Carlo-based simulation toolbox. An optimal design was chosen after several iterations starting from the initial design. To cover a wide region of interest, this unit design was extended in a modular fashion, while respecting mechanical restrictions as well as the need for dense distribution of sources and detectors.

Standard techniques for detection of thyroid cancer (ultrasound screening and fine-needle aspiration biopsy) have limited sensitivity and specificity, leading to a very large number of unnecessary thyroid extraction surgeries. With the aim of improving diagnosis, hybrid diffuse optics and ultrasound were used on nodules patients to obtain tissue hemodynamic information. Nodules rated 4A or 4B in the thyroid imaging reporting and data system (TI-RADS) are of clinical relevance and were classified using a logistic regression model built on our results. Fourteen benign and four malignant nodules were classified with a sensitivity of 100% and specificity of 77%.

We have developed a forward-scattered high-resolution Raman-Computed Tomography (R-CT) technique that relies on true CT principles rather than diffuse tomography. The in-house built optical system rotates around the sample while Raman spectra are measured from different angles by multiplexing on the 2D CCD. Forward scattered photons are largely in the far-field to reject diffuse photons. We show that R-CT provides both, spatial and molecular information in diffusely scattering phantoms. Data reconstruction is performed using backprojection and multivariate regression analysis to provide a 3D chemical contrast. We finally present preliminary results on rodent limbs. The R-CT technique developed offers a potential technique to improve the characterisation of tissues and drug efficiency in rodent limbs at the molecular level as it offers morphological information that is not available in conventional transmission Raman spectroscopy.

We report neural network-based rapid reconstruction of swept-source OCT (SS-OCT) images using undersampled spectral data. We trained and blindly tested a deep neural network using mouse embryo samples imaged by an SS-OCT system. Using >3-fold undersampled spectral data per A-line, the trained neural network can blindly remove spatial aliasing artifacts due to spectral undersampling, presenting a very good match to the images reconstructed using the full spectral data. This method can be integrated with various swept-source or spectral domain OCT systems to potentially improve the 3D imaging speed without a sacrifice in resolution or signal-to-noise of the reconstructed images.

Ageing affects both the structure and the function of the skin. This results in degradation of physiological functions and biophysical properties of organs and tissues putting an alarming economic and social burden on governments and healthcare services. Any significant alteration of skin’s mechanical properties that could affect its structural integrity and barrier function has the potential to cause health effects. With age, the collagen and elastin fibre networks in the dermis become compacted leading to reduction in dermis thickness. Skin photoaging leads to solar elastosis or disintegration of elastic fibers and accumulation of amorphous, thick ended, fragmented elastic fibers. The skin aging process is not yet fully understood.