Generalizability of Deep Neural Networks for Vertical Cup-to-Disc Ratio Estimation in Ultra-Widefield and Smartphone-Based Fundus Images.

Purpose: To develop and validate a deep learning system (DLS) for estimation of vertical cup-to-disc ratio (vCDR) in ultra-widefield (UWF) and smartphone-based fundus images. Methods: A DLS consisting of two sequential convolutional neural networks (CNNs) to delineate optic disc (OD) and optic cup (OC) boundaries was developed using 800 standard fundus images from the public REFUGE data set. The CNNs were tested on 400 test images from the REFUGE data set and 296 UWF and 300 smartphone-based images from a teleophthalmology clinic. vCDRs derived from the delineated OD/OC boundaries were compared with optometrists' annotations using mean absolute error (MAE). Subgroup analysis was conducted to study the impact of peripapillary atrophy (PPA), and correlation study was performed to investigate potential correlations between sectoral CDR (sCDR) and retinal nerve fiber layer (RNFL) thickness. Results: The system achieved MAEs of 0.040 (95% CI, 0.037-0.043) in the REFUGE test images, 0.068 (95% CI, 0.061-0.075) in the UWF images, and 0.084 (95% CI, 0.075-0.092) in the smartphone-based images. There was no statistical significance in differences between PPA and non-PPA images. Weak correlation (r = -0.4046, P < 0.05) between sCDR and RNFL thickness was found only in the superior sector. Conclusions: We developed a deep learning system that estimates vCDR from standard, UWF, and smartphone-based images. We also described anatomic peripapillary adversarial lesion and its potential impact on OD/OC delineation. Translational Relevance: Artificial intelligence can estimate vCDR from different types of fundus images and may be used as a general and interpretable screening tool to improve community reach for diagnosis and management of glaucoma.

Coarse-to-fine visual representation learning for medical images via class activation maps.

The value of coarsely labeled datasets in learning transferable representations for medical images is investigated in this work. Compared to fine labels which require meticulous effort to annotate, coarse labels can be acquired at a significantly lower cost and can provide useful training signals for data-hungry deep neural networks. We consider coarse labels in the form of binary labels differentiating a normal (healthy) image from an abnormal (diseased) image and propose CAMContrast, a two-stage representation learning framework for medical images. Using class activation maps, CAMContrast makes use of the binary labels to generate heatmaps as positive views for contrastive representation learning. Specifically, the learning objective is optimized to maximize the agreement within fixed crops of image-heatmap pair to learn fine-grained representations that are generalizable to different downstream tasks. We empirically validate the transfer learning performance of CAMContrast on several public datasets, covering classification and segmentation tasks on fundus photographs and chest X-ray images. The experimental results showed that our method outperforms other self-supervised and supervised pretrain methods in terms of data efficiency and downstream performance.

Improving single pixel imaging performance in high noise condition by under-sampling.

Single-pixel imaging could be a superior solution for imaging applications where the detector array is very expensive or not even available. Sampling order, sampling ratio, noise and type of transforms affect the quality of the reconstructed image. Here, we compare the performance of single pixel imaging (SPI) with Hadamard transform (HT) and discrete cosine transform (DCT) in the presence of noise. The trade-off between adding image information and adding noise in each coefficient measurement results in an optimum number of measurements for reconstruction image quality. In addition, DCT shows higher image quality with fewer measurements than HT does. We then demonstrate our SPI with optimum sampling strategy for a large set of images and lab experiments and finally put forward a quality control technique, which is corroborated by the practical experiments. Our results suggest a practical approach for SPI to improve the speed and achieve the highest possible image quality.

Functionalized MoS2 Nanosheets as Multi-Gene Delivery Vehicles for In Vivo Pancreatic Cancer Therapy.

Transition metal dichalcogenides (TMDCs) are categorized as novel two-dimensional (2D) nanomaterials with unique physical and chemical properties, bearing varied applications in medical and materials sciences. However, only a few works report the application of TMDCs for gene therapy in cancer treatment. Here, we engineer a multi-gene delivery system based on functionalized monolayer MoS2, which can co-deliver HDAC1 and KRAS small interfering RNAs (siRNAs) to Panc-1 cancer cells for combinational cancer therapy. The synergistic effect of gene silencing therapy and NIR phototherapy is demonstrated by inhibition of both genes, in vitro cell growth rate, and in vivo tumor volume growth rate, exemplifying pre-eminent anticancer efficacy. This anti-tumor effect is a result of the photothermal effect of MoS2 induced by NIR excitation and inactivation of HDAC1 and KRAS genes, which consequently bring about apoptosis, inhibit migration, and induce cell cycle arrest in the treated Panc-1 cells. Moreover, good biocompatibility and reduced cytotoxicity of MoS2-based nanocarriers enable their metabolism within in vitro and in vivo mouse models over a prolonged duration without any evident ill-effects. In summary, we demonstrate the promising potential of low-toxicity, functionalized MoS2 nanocarriers as a biocompatible gene delivery system for in vivo pancreatic adenocarcinoma therapy.

Enhancement of early cervical cancer diagnosis with epithelial layer analysis of fluorescence lifetime images.

This work reports the use of layer analysis to aid the fluorescence lifetime diagnosis of cervical intraepithelial neoplasia (CIN) from H&E stained cervical tissue sections. The mean and standard deviation of lifetimes in single region of interest (ROI) of cervical epithelium were previously shown to correlate to the gold standard histopathological classification of early cervical cancer. These previously defined single ROIs were evenly divided into layers for analysis. A 10-layer model revealed a steady increase in fluorescence lifetime from the inner to the outer epithelial layers of healthy tissue sections, suggesting a close association with cellular maturity. The shorter lifetime and minimal lifetime increase towards the epithelial surface of CIN-affected regions are in good agreement with the absence of cellular maturation in CIN. Mean layer lifetimes in the top-half cervical epithelium were used as feature vectors for extreme learning machine (ELM) classifier discriminations. It was found that the proposed layer analysis technique greatly improves the sensitivity and specificity to 94.6% and 84.3%, respectively, which can better supplement the traditional gold standard cervical histopathological examinations.

Quantitative photothermal phase imaging of red blood cells using digital holographic photothermal microscope.

Photothermal microscopy (PTM), a noninvasive pump-probe high-resolution microscopy, has been applied as a bioimaging tool in many biomedical studies. PTM utilizes a conventional phase contrast microscope to obtain highly resolved photothermal images. However, phase information cannot be extracted from these photothermal images, as they are not quantitative. Moreover, the problem of halos inherent in conventional phase contrast microscopy needs to be tackled. Hence, a digital holographic photothermal microscopy technique is proposed as a solution to obtain quantitative phase images. The proposed technique is demonstrated by extracting phase values of red blood cells from their photothermal images. These phase values can potentially be used to determine the temperature distribution of the photothermal images, which is an important study in live cell monitoring applications.

Quantitative diagnosis of cervical neoplasia using fluorescence lifetime imaging on haematoxylin and eosin stained tissue sections.

The use of conventional fluorescence microscopy for characterizing tissue pathological states is limited by overlapping spectra and the dependence on excitation power and fluorophore concentration. Fluorescence lifetime imaging microscopy (FLIM) can overcome these limitations due to its insensitivity to fluorophore concentration, excitation power and spectral similarity. This study investigates the diagnosis of early cervical cancer using FLIM and a neural network extreme learning machine classifier. A concurrently high sensitivity and specificity of 92.8% and 80.2%, respectively, were achieved. The results suggest that the proposed technique can be used to supplement the traditional histopathological examination of early cervical cancer.

Colorimetric surface plasmon resonance imaging (SPRI) biosensor array based on polarization orientation.

A colorimetric surface plasmon resonance (SPR) imaging biosensor array based on polarization orientation rotation is presented in this paper. It measures the spectral characteristic variations caused by the steep phase difference between the p- and s-polarization occurring at surface plasmon excitation. It provides one-order of magnitude sensor resolution improvement comparing to existing phase-sensitive SPR imaging sensors and the two-dimensional (2D) sensing capability of the imaging sensor enables multiplex, high throughput array based simultaneous detection for a range of different bio-molecular interactions. Experiments on the binding interactions detection between anti-bovine serum albumin (anti-BSA) and BSA antigen have been performed. All binding interactions occurred at 5×4 protein array were real-time monitored simultaneously. A sensor resolution of 8.26ng/ml (125pM) has been demonstrated, which is one-order of magnitude (12 times) better than the detection limit reported by existing phase-sensitive SPR imaging sensors in the literature, while no time-consuming phase modulation and phase extraction processes are required. Furthermore, the optical colorimetric image read-out of the sensor is easy to be identified by the end users comparing to conventional intensity or phase information. The colorimetric SPR imaging biosensor array can find promising potential applications in high throughput clinical disease diagnosis, protein biomarkers screening and drug screening.

Dual-window dual-bandwidth spectroscopic optical coherence tomography metric for qualitative scatterer size differentiation in tissues.

This study investigates the autocorrelation bandwidths of dual-window (DW) optical coherence tomography (OCT) k-space scattering profile of different-sized microspheres and their correlation to scatterer size. A dual-bandwidth spectroscopic metric defined as the ratio of the 10% to 90% autocorrelation bandwidths is found to change monotonically with microsphere size and gives the best contrast enhancement for scatterer size differentiation in the resulting spectroscopic image. A simulation model supports the experimental results and revealed a tradeoff between the smallest detectable scatterer size and the maximum scatterer size in the linear range of the dual-window dual-bandwidth (DWDB) metric, which depends on the choice of the light source optical bandwidth. Spectroscopic OCT (SOCT) images of microspheres and tonsil tissue samples based on the proposed DWDB metric showed clear differentiation between different-sized scatterers as compared to those derived from conventional short-time Fourier transform metrics. The DWDB metric significantly improves the contrast in SOCT imaging and can aid the visualization and identification of dissimilar scatterer size in a sample. Potential applications include the early detection of cell nuclear changes in tissue carcinogenesis, the monitoring of healing tendons, and cell proliferation in tissue scaffolds.

Multiplex spectral surface plasmon resonance imaging (SPRI) sensor based on the polarization control scheme.

A two-dimensional (2D) spectral SPR sensor based on a polarization control scheme is reported in this paper. The polarization control configuration converts the phase difference between p- and s- polarization occurring at surface plasmon resonance (SPR) into corresponding color responses in spectral SPR images. A sensor resolution of 2.7 x 10(-6) RIU has been demonstrated, which corresponds to more than one order of magnitude resolution improvement (26 times) comparing to existing 2D spectral SPR sensors. Multiplex array detection has also been demonstrated with the spectral SPR imaging sensor. In a 8 x 4 sensor array, 32 samples with different refractive index values were monitored simultaneously. Detection on bovine serum albumin (BSA) antigen-antibody binding further demonstrated the multiplex detection capability of the 2D spectral SPR sensor for bio-molecular interactions. The detection limit is found to be 21 ng/ml, which is 36 times better than the detection limit previously reported by phase imaging SPR sensors. In light of the advantages of high sensitivity, 2D multiplex detection and real-time response, the spectral SPR imaging sensor can find promising applications in rapid, high throughput, non-labeling and multiplex detection of protein array for proteomics studies, biomarker screening, disease prognosis, and drug discovery.

Concentration dependence of gold nanoshells on the enhancement of optical coherence tomography images: a quantitative study.

The effective use of gold nanoshells as a contrast agent for optical coherence tomography (OCT) may be hampered by the delivery of a wrong dose to tissue that results in unwanted signal attenuation. In this study we examine how changes in mu(s) due to concentration variations affect the OCT image and then define a dosing range that would result in appropriate scattering coefficient, mu(s), to maintain an acceptable signal attenuation level. Our results show that an increase in sample mu(s) not only enhances the OCT signal near the surface but also attenuates the signal deeper into the sample. We synthesized gold nanoshells with an 81 nm radius silica core and 23 nm shell thickness and found that a concentration range of 5.6 x 10(9)

Fluorescence lifetime discrimination using expectation-maximization algorithm with joint deconvolution.

The fluorescence lifetime technique offers an effective way to resolve fluorescent components with overlapping emission spectra. The presence of multiple fluorescent components in biological compounds can hamper their discrimination. The conventional method based on the nonlinear least-squares technique is unable to consistently determine the correct number of fluorescent components in a fluorescence decay profile. This can limit the applications of the fluorescence lifetime technique in biological assays and diagnoses where more than one fluorescent component is typically encountered. We describe the use of an expectation-maximization (EM) method with joint deconvolution to estimate the fluorescence decay parameters, and the Bayesian information criterion (BIC) to accurately determine the number of fluorescent components. A comprehensive simulation and experimental study is carried out to compare the performance and accuracy of the proposed method. The results show that the EM-BIC method is able to accurately identify the correct number of fluorescent components in samples with weakly fluorescing components.

Fluorescence detection of bladder cancer using urine cytology.

Bladder cancer is the fourth most common malignant disease worldwide, accounting for 4% of all cancer cases. In Singapore, it is the ninth most common form of cancer. The high mortality rate in bladder cancer can be reduced by early treatment following pre-cancerous screening. Currently, the gold standard for screening bladder tumors is histological examination of biopsy specimens, which is both invasive and time-consuming. In this study, ex vivo urine fluorescence cytology was investigated to offer an alternative timely and biopsy-free means for detecting bladder cancers. Sediments in patient urine samples were extracted and incubated with a novel photosensitizer, hypericin. Laser confocal microscopy was used to capture the fluorescence images at an excitation wavelength of 488 nm. Images were subsequently processed to single out the exfoliated bladder cancer cells from the other cells based on the cellular size. Intensity histograms of each targeted cell and feature vectors, derived from the histogram moments, were used to represent each sample. A difference in the distribution of the feature vectors of normal and low-grade cancerous bladder cancer cells were observed. A diagnostic algorithm for discriminating between normal and low-grade cancerous cells is elucidated in this report. This study suggests that the fluorescence intensity profiles of hypericin in bladder cells can potentially provide an automated quantitative means of early bladder cancer diagnosis.