Handcrafted vs. non-handcrafted features for computer vision classification

Near-infrared (NIR) activatable fluorescent probes have been considered to be the effective edge tools for the investigation of cell biology and disease diagnosis because of their outstanding advantages. Related genes involved in tumor genesis and progression regulate the overexpression of certain enzymes. Owing to the distinctive characteristics of quick reaction time and favorable pharmacokinetics, enzyme-reactive NIR optical probes have shown great potential in the diagnosis of tumorigenesis and in image-guided intraoperative surgeries with high signal-to-noise ratios. In this review, we mainly summarize the latest advancements in enzyme-reactive NIR fluorescent probes from design strategy to biomedical application. Specifically, some challenges and prospects in this field are presented at the end of the review, which will be beneficial to innovatively construct new multifunctional fluorescent probes and actively promote their clinical transformation in the future.

Activity‐Based NIR Enzyme Fluorescent Probes for the Diagnosis of Tumors and Image‐Guided Surgery

The twisted intramolecular charge transfer (TICT) mechanism has guided the development of numerous bright and sensitive fluorophores. This review briefly overviews the history of establishing the TICT mechanism, and systematically summarizes the molecular design strategies in modulating the TICT tendency of various organic fluorophores towards different applications, along with key milestone studies and representative examples. Additionally, we also succinctly review the twisted intramolecular charge shuttle (TICS) and twists during photoinduced electron transfer (PET), and compare their similarities and differences with TICT, with emphasis on understanding the structure–property relationships between the twisted geometries and how they can directly affect the fluorescence of the molecules. Such structure–property relationships presented herein will greatly aid the rational development of fluorophores that involve molecular twisting in the excited state.

Twisted intramolecular charge transfer (TICT) and twists beyond TICT: from mechanisms to rational designs of bright and sensitive fluorophores

Fluorescence imaging has made tremendous inroads toward understanding the complexity of biological systems, but in vivo deep-tissue imaging remains a great challenge due to the optical opacity of biological tissue. Recent improvements in laser and detector manufacturing have allowed the expansion of nonlinear and linear fluorescence imaging to the underexplored "tissue-transparent" second near-infrared (NIR-II; 1000-1700 nm) window, opening up new opportunities for optical access deep inside opaque tissue. Molecular fluorophores have historically played a major role in fluorescence bioimaging. It is increasingly important to design new molecular fluorophores to fully unlock the potential of NIR-II imaging techniques. In this outlook, we give an overview of the novel molecular fluorophores developed for deep-tissue bioimaging in the past five years and discuss their pros and cons in applications. Guidelines for designing new molecular fluorophores with the desirable properties are also provided.

Molecular Fluorophores for Deep-Tissue Bioimaging.

Currently, the application of deep learning in crop disease classification is one of the active areas of research for which an image dataset is required. Eggplant (Solanum melongena) is one of the important crops, but it is susceptible to serious diseases which hinder its production. Surprisingly, so far no dataset is available for the diseases in this crop. The unavailability of the dataset for these diseases motivated the authors to create a standard dataset in laboratory and field conditions for five major diseases. Pre-trained Visual Geometry Group 16 (VGG16) architecture has been used and the images have been converted to other color spaces namely Hue Saturation Value (HSV), YCbCr and grayscale for evaluation. Results show that the dataset created with RGB and YCbCr images in field condition was promising with a classification accuracy of 99.4%. The dataset also has been evaluated with other popular architectures and compared. In addition, VGG16 has been used as feature extractor from 8th convolution layer and these features have been used for classifying diseases employing Multi-Class Support Vector Machine (MSVM). The analysis depicted an equivalent or in some cases produced better accuracy. Possible reasons for variation in interclass accuracy and future direction have been discussed.

/pdf/disease-classification-in-eggplant-using-pre-trained-vgg16-3rzow342oz.pdf

Disease Classification in Eggplant Using Pre-trained VGG16 and MSVM.

Photoinduced electron transfer (PET) is one of the most important mechanisms for developing fluorescent probes and biosensors. Quantitative prediction of the quantum yields of these probes and sensors is crucial to accelerate the rational development of novel PET-based functional materials. Herein, we developed a general descriptor (ΔE) for predicting the quantum yield of PET probes, with a threshold value of ∼0.6 eV. When ΔE ∼0.6 eV, the quantum yield is high because of the inhibition of PET. This simple yet effective descriptor is applicable to a wide range of fluorophores, such as BODIPY, fluorescein, rhodamine, and Si-rhodamine. This ΔE descriptor enables us not only to establish new applications for existing PET probes but also to quantitatively design novel PET-based fluorophores for wash-free bioimaging and AIEgen development.

A General Descriptor ΔE Enables the Quantitative Development of Luminescent Materials Based on Photoinduced Electron Transfer

Convolutional Neural Networks (CNNs) are powerful models that achieve impressive results for image classification. In addition, pre-trained CNNs are also useful for other computer vision tasks as generic feature extractors. This paper aims to gain insight into the feature aspect of CNN and demonstrate other uses of CNN features. Our results show that CNN feature maps can be used with Random Forests and SVM to yield classification results that outperforms the original CNN. A CNN that is less than optimal (e.g. not fully trained or overfitting) can also extract features for Random Forest/SVM that yield competitive classification accuracy. In contrast to the literature which uses the top-layer activations as feature representation of images for other tasks, using lower-layer features can yield better results for classification.

Feature Representation in Convolutional Neural Networks

Multivariable calculus is one of the core subjects which engineering students study at university. Students may end up memorizing formulae in multivariable calculus due to the lack of ability in visualizing 3D surfaces, and hence not gain much intuition for the subject. We developed an in-house virtual reality (VR) application with the intention for students to visualize concepts in Multivariable Calculus. In order to evaluate the effectiveness of the VR application (which we term as the treatment), we performed a blinded randomized controlled trial with n = 312 students, where we divided them into a control group of n
 CO
 = 187 students and treatment group of n
 TR
 = 125 students. We gave both groups of students a test immediately after the treatment, as well as asking them to fill in anonymous survey questions using a Likert scale. Our findings show that students perform worse on some questions after using the VR application, and for some other questions students have similar performance to the treatment group. We hypothesize some reasons why this is so, opening the door for future research. We also give recommendations for future developers of VR applications.

/pdf/impact-of-virtual-reality-on-the-visualization-of-partial-146ul7lxak.pdf

Impact of Virtual Reality on the Visualization of Partial Derivatives in a Multivariable Calculus Class

Random projections are used to estimate parameters of interest in large scale data sets by projecting data into a lower dimensional space. Some parameters of interest between pairs of vectors are the Euclidean distance and the inner product, while parameters of interest for the whole data set could be its singular values or singular vectors. We show how we can borrow an idea from Monte Carlo integration by using control variates to reduce the variance of the estimates of Euclidean distances and inner products by storing marginal information of our data set. We demonstrate this variance reduction through experiments on synthetic data as well as the colon and kos datasets. We hope that this inspires future work which incorporates control variates in further random projection applications.

Random Projections with Control Variates

Random projection is a dimension reduction technique which can be used to estimate Euclidean distances, inner products, angles [9], or even \(l_p\) distances (for even p) [10] between pairs of high dimensional vectors. We extend the work of Li [9] and our prior work [7] to show how marginal information, principal components, and control variates can be used with the multivariate normal distribution to improve the accuracy of the inner product estimate of vectors. We call our method COntrol Variates For Estimation via First Eigenvectors (COVFEFE). We demonstrate the results of COVFEFE on the Arcene and MNIST datasets.

Keegan Kang

Papers

A General Descriptor ΔE Enables the Quantitative Development of Luminescent Materials Based on Photoinduced Electron Transfer

Feature Representation in Convolutional Neural Networks

Impact of Virtual Reality on the Visualization of Partial Derivatives in a Multivariable Calculus Class

Random Projections with Control Variates

Using the Multivariate Normal to Improve Random Projections