Epiretinal Membrane Detection at the Ophthalmologist Level using Deep Learning of Optical Coherence Tomography

PURPOSE

Previous deep learning studies on optical coherence tomography (OCT) mainly focused on diabetic retinopathy and age-related macular degeneration. We proposed a deep learning model that can identify epiretinal membrane (ERM) in OCT with ophthalmologist-level performance.

DESIGN

Cross-sectional study.

PARTICIPANTS

A total of 3,618 central fovea cross section OCT images from 1,475 eyes of 964 patients.

METHODS

We retrospectively collected 7,652 OCT images from 1,197 patients. From these images, 2,171 were normal and 1,447 were ERM OCT. A total of 3,141 OCT images was used as training dataset and 477 images as testing dataset. DL algorithm was used to train the interpretation model. Diagnostic results by four board-certified non-retinal specialized ophthalmologists on the testing dataset were compared with those generated by the DL model.

MAIN OUTCOME MEASURES

We calculated for the derived DL model the following characteristics: sensitivity, specificity, F1 score and area under curve (AUC) of the receiver operating characteristic (ROC) curve. These were calculated according to the gold standard results which were parallel diagnoses of the retinal specialist. Performance of the DL model was finally compared with that of non-retinal specialized ophthalmologists.

RESULTS

Regarding the diagnosis of ERM in OCT images, the trained DL model had the following characteristics in performance: sensitivity: 98.7%, specificity: 98.0%, and F1 score: 0.945. The accuracy on the training dataset was 99.7% (95% CI: 99.4 - 99.9%), and for the testing dataset, diagnostic accuracy was 98.1% (95% CI: 96.5 - 99.1%). AUC of the ROC curve was 0.999. The DL model slightly outperformed the average non-retinal specialized ophthalmologists.

CONCLUSIONS

An ophthalmologist-level DL model was built here to accurately identify ERM in OCT images. The performance of the model was slightly better than the average non-retinal specialized ophthalmologists. The derived model may play a role to assist clinicians to promote the efficiency and safety of healthcare in the future.

Evaluation of Deep Learning Neural Networks for Surface Roughness Prediction Using Vibration Signal Analysis

The use of surface roughness (Ra) to indicate product quality in the milling process in an intelligent monitoring system applied in-process has been developing. From the considerations of convenient installation and cost-effectiveness, accelerator vibration signals combined with deep learning predictive models for predicting surface roughness is a potential tool. In this paper, three models, namely, Fast Fourier Transform-Deep Neural Networks (FFT-DNN), Fast Fourier Transform Long Short Term Memory Network (FFT-LSTM), and one-dimensional convolutional neural network (1-D CNN), are used to explore the training and prediction performances. Feature extraction plays an important role in the training and predicting results. FFT and the one-dimensional convolution filter, known as 1-D CNN, are employed to extract vibration signals’ raw data. The results show the following: (1) the LSTM model presents the temporal modeling ability to achieve a good performance at higher Ra value and (2) 1-D CNN, which is better at extracting features, exhibits highly accurate prediction performance at lower Ra ranges. Based on the results, vibration signals combined with a deep learning predictive model could be applied to predict the surface roughness in the milling process. Based on this experimental study, the use of prediction of the surface roughness via vibration signals using FFT-LSTM or 1-D CNN is recommended to develop an intelligent system.

Implementation of a parallel protein structure alignment service on cloud

Protein structure alignment has become an important strategy by which to identify evolutionary relationships between protein sequences. Several alignment tools are currently available for online comparison of protein structures. In this paper, we propose a parallel protein structure alignment service based on the Hadoop distribution framework. This service includes a protein structure alignment algorithm, a refinement algorithm, and a MapReduce programming model. The refinement algorithm refines the result of alignment. To process vast numbers of protein structures in parallel, the alignment and refinement algorithms are implemented using MapReduce. We analyzed and compared the structure alignments produced by different methods using a dataset randomly selected from the PDB database. The experimental results verify that the proposed algorithm refines the resulting alignments more accurately than existing algorithms. Meanwhile, the computational performance of the proposed service is proportional to the number of processors used in our cloud platform.

A Smartphone-Based Application for Scale Pest Detection Using Multiple-Object Detection Methods

Taiwan’s economy mainly relies on the export of agricultural products. If even the suspicion of a pest is found in the crop products after they are exported, not only are the agricultural products returned but the whole batch of crops is destroyed, resulting in extreme crop losses. The species of mealybugs, Coccidae, and Diaspididae, which are the primary pests of the scale insect in Taiwan, can not only lead to serious damage to the plants but also severely affect agricultural production. Hence, to recognize the scale pests is an important task in Taiwan’s agricultural field. In this study, we propose an AI-based pest detection system for solving the specific issue of detection of scale pests based on pictures. Deep-learning-based object detection models, such as faster region-based convolutional networks (Faster R-CNNs), single-shot multibox detectors (SSDs), and You Only Look Once v4 (YOLO v4), are employed to detect and localize scale pests in the picture. The experimental results show that YOLO v4 achieved the highest classification accuracy among the algorithms, with 100% in mealybugs, 89% in Coccidae, and 97% in Diaspididae. Meanwhile, the computational performance of YOLO v4 has indicated that it is suitable for real-time application. Moreover, the inference results of the YOLO v4 model further help the end user. A mobile application using the trained scale pest recognition model has been developed to facilitate pest identification in farms, which is helpful in applying appropriate pesticides to reduce crop losses.

GPU-Accelerated GLRLM Algorithm for Feature Extraction of MRI

The gray level run length matrix (GLRLM) whose entries are statistics recording distribution and relationship of images pixels is a widely used method for extracting statistical features for medical images, e.g., magnetic resonance (MR) images. Recently these features are usually employed in some artificial neural networks to identify and distinguish texture patterns. But GLRLM construction and features extraction are tedious and computationally intensive while the images are too big with high resolution, or there are too many small or intermediate Regions of Interest (ROI) to process in a single image, which makes the preprocess a time consuming stage. Hence, it is of great importance to accelerate the procedure which is nowadays possible with the rapid development of massively parallel Graphics Processing Unit, i.e. the GPU computing technology. In this article, we propose a new paradigm based on mature parallel primitives for generating GLRLMs and extracting multiple features for many ROIs simultaneously in a single image. Experiments show that such a paradigm is easy to implement and offers an acceleration over 5 fold increase in speed than an optimized serial counterpart.

CUDA ClustalW: An efficient parallel algorithm for progressive multiple sequence alignment on Multi-GPUs

For biological applications, sequence alignment is an important strategy to analyze DNA and protein sequences. Multiple sequence alignment is an essential methodology to study biological data, such as homology modeling, phylogenetic reconstruction and etc. However, multiple sequence alignment is a NP-hard problem. In the past decades, progressive approach has been proposed to successfully align multiple sequences by adopting iterative pairwise alignments. Due to rapid growth of the next generation sequencing technologies, a large number of sequences can be produced in a short period of time. When the problem instance is large, progressive alignment will be time consuming. Parallel computing is a suitable solution for such applications, and GPU is one of the important architectures for contemporary parallel computing researches. Therefore, we proposed a GPU version of ClustalW v2.0.11, called CUDA ClustalW v1.0, in this work. From the experiment results, it can be seen that the CUDA ClustalW v1.0 can achieve more than 33× speedups for overall execution time by comparing to ClustalW v2.0.11.

Cloud computing for protein-ligand binding site comparison

The proteome-wide analysis of protein-ligand binding sites and their interactions with ligands is important in structure-based drug design and in understanding ligand cross reactivity and toxicity. The well-known and commonly used software, SMAP, has been designed for 3D ligand binding site comparison and similarity searching of a structural proteome. SMAP can also predict drug side effects and reassign existing drugs to new indications. However, the computing scale of SMAP is limited. We have developed a high availability, high performance system that expands the comparison scale of SMAP. This cloud computing service, called Cloud-PLBS, combines the SMAP and Hadoop frameworks and is deployed on a virtual cloud computing platform. To handle the vast amount of experimental data on protein-ligand binding site pairs, Cloud-PLBS exploits the MapReduce paradigm as a management and parallelizing tool. Cloud-PLBS provides a web portal and scalability through which biologists can address a wide range of computer-intensive questions in biology and drug discovery.

Liposomal n-butylidenephthalide protects the drug from oxidation and enhances its antitumor effects in glioblastoma multiforme

Background

The natural compound n-butylidenephthalide (BP) can pass through the blood–brain barrier to inhibit the growth of glioblastoma multiforme tumors. However, BP has an unstable structure that reduces its antitumor activity and half-life in vivo.

Objective

The aim of this study is to design a drug delivery system to encapsulate BP to enhance its efficacy by improving its protection and delivery.

Methods

To protect its structural stability against protein-rich and peroxide solutions, BP was encapsulated into a lipo-PEG-PEI complex (LPPC). Then, the cytotoxicity of BP/LPPC following preincubation in protein-rich, acid/alkaline, and peroxide solutions was analyzed by MTT. Cell uptake of BP/LPPC was also measured by confocal microscopy. The therapeutic effects of BP/LPPC were analyzed in xenograft mice following intratumoral and intravenous injections.

Results

When BP was encapsulated in LPPC, its cytotoxicity was maintained following preincubation in protein-rich, acid/alkaline, and peroxide solutions. The cytotoxic activity of encapsulated BP was higher than that of free BP (~4.5- to 8.5-fold). This increased cytotoxic activity of BP/LPPC is attributable to its rapid transport across the cell membrane. In an animal study, a subcutaneously xenografted glioblastoma multiforme mouse that was treated with BP by intratumoral and intravenous administration showed inhibited tumor growth. The same dose of BP/LPPC was significantly more effective in terms of tumor inhibition.

Conclusion

LPPC encapsulation technology is able to protect BP’s structural stability and enhance its antitumor effects, thus providing a better tool for use in cancer therapy.

GPU-based cloud service for Smith-Waterman algorithm using frequency distance filtration scheme

As the conventional means of analyzing the similarity between a query sequence and database sequences, the Smith-Waterman algorithm is feasible for a database search owing to its high sensitivity. However, this algorithm is still quite time consuming. CUDA programming can improve computations efficiently by using the computational power of massive computing hardware as graphics processing units (GPUs). This work presents a novel Smith-Waterman algorithm with a frequency-based filtration method on GPUs rather than merely accelerating the comparisons yet expending computational resources to handle such unnecessary comparisons. A user friendly interface is also designed for potential cloud server applications with GPUs. Additionally, two data sets, H1N1 protein sequences (query sequence set) and human protein database (database set), are selected, followed by a comparison of CUDA-SW and CUDA-SW with the filtration method, referred to herein as CUDA-SWf. Experimental results indicate that reducing unnecessary sequence alignments can improve the computational time by up to 41%. Importantly, by using CUDA-SWf as a cloud service, this application can be accessed from any computing environment of a device with an Internet connection without time constraints.

Health Care Competition and Antibiotic Use in Taiwan

Antibiotic resistance, a negative externality of antibiotic use, is a growing threat to public health. Health care competition may encourage antibiotic use because receiving an antibiotic is a form of ‘quality’ for many patients. This paper examines the effect of market concentration on antibiotic use in a large, nationally‐representative data set from Taiwan. Moving from the 75th percentile to the 25th percentile of market concentration is associated with 6.6 per cent greater antibiotic use. We control for leading market‐level confounds, including population density and community health. We also show that the correlation is robust using fixed effects for patients, physicians and diagnoses. We document the correlation between antibiotic use and patient retention, which suggests a mechanism for this result. Finally, we show that strict regulation of antibiotics reduces but does not eliminate the effect of competition on antibiotic use.

MGUPGMA: A Fast UPGMA Algorithm With Multiple Graphics Processing Units Using NCCL

A phylogenetic tree is a visual diagram of the relationship between a set of biological species. The scientists usually use it to analyze many characteristics of the species. The distance-matrix methods, such as Unweighted Pair Group Method with Arithmetic Mean and Neighbor Joining, construct a phylogenetic tree by calculating pairwise genetic distances between taxa. These methods have the computational performance issue. Although several new methods with high-performance hardware and frameworks have been proposed, the issue still exists. In this work, a novel parallel Unweighted Pair Group Method with Arithmetic Mean approach on multiple Graphics Processing Units is proposed to construct a phylogenetic tree from extremely large set of sequences. The experimental results present that the proposed approach on a DGX-1 server with 8 NVIDIA P100 graphic cards achieves approximately 3-fold to 7-fold speedup over the implementation of Unweighted Pair Group Method with Arithmetic Mean on a modern CPU and a single GPU, respectively.

Local alignment tool based on Hadoop framework and GPU architecture

With the rapid growth of next generation sequencing technologies, such as Slex, more and more data have been discovered and published. To analyze such huge data the computational performance is an important issue. Recently, many tools, such as SOAP, have been implemented on Hadoop and GPU parallel computing architectures. BLASTP is an important tool, implemented on GPU architectures, for biologists to compare protein sequences. To deal with the big biology data, it is hard to rely on single GPU. Therefore, we implement a distributed BLASTP by combining Hadoop and multi-GPUs. The experimental results present that the proposed method can improve the performance of BLASTP on single GPU, and also it can achieve high availability and fault tolerance.

Open reading frame phylogenetic analysis on the cloud

Phylogenetic analysis has become essential in researching the evolutionary relationships between viruses. These relationships are depicted on phylogenetic trees, in which viruses are grouped based on sequence similarity. Viral evolutionary relationships are identified from open reading frames rather than from complete sequences. Recently, cloud computing has become popular for developing internet-based bioinformatics tools. Biocloud is an efficient, scalable, and robust bioinformatics computing service. In this paper, we propose a cloud-based open reading frame phylogenetic analysis service. The proposed service integrates the Hadoop framework, virtualization technology, and phylogenetic analysis methods to provide a high-availability, large-scale bioservice. In a case study, we analyze the phylogenetic relationships among Norovirus. Evolutionary relationships are elucidated by aligning different open reading frame sequences. The proposed platform correctly identifies the evolutionary relationships between members of Norovirus.

Cloud computing-based TagSNP selection algorithm for human genome data

Single nucleotide polymorphisms (SNPs) play a fundamental role in human genetic variation and are used in medical diagnostics, phylogeny construction, and drug design. They provide the highest-resolution genetic fingerprint for identifying disease associations and human features. Haplotypes are regions of linked genetic variants that are closely spaced on the genome and tend to be inherited together. Genetics research has revealed SNPs within certain haplotype blocks that introduce few distinct common haplotypes into most of the population. Haplotype block structures are used in association-based methods to map disease genes. In this paper, we propose an efficient algorithm for identifying haplotype blocks in the genome. In chromosomal haplotype data retrieved from the HapMap project website, the proposed algorithm identified longer haplotype blocks than an existing algorithm. To enhance its performance, we extended the proposed algorithm into a parallel algorithm that copies data in parallel via the Hadoop MapReduce framework. The proposed MapReduce-paralleled combinatorial algorithm performed well on real-world data obtained from the HapMap dataset; the improvement in computational efficiency was proportional to the number of processors used.

Automated Classification of Blood Loss from Transurethral Resection of the Prostate Surgery Videos Using Deep Learning Technique

Transurethral resection of the prostate (TURP) is a surgical removal of obstructing prostate tissue. The total bleeding area is used to determine the performance of the TURP surgery. Although the traditional method for the detection of bleeding areas provides accurate results, it cannot detect them in time for surgery diagnosis. Moreover, it is easily disturbed to judge bleeding areas for experienced physicians because a red light pattern arising from the surgical cutting loop often appears on the images. Recently, the automatic computer-aided technique and artificial intelligence deep learning are broadly used in medical image recognition, which can effectively extract the desired features to reduce the burden of physicians and increase the accuracy of diagnosis. In this study, we integrated two state-of-the-art deep learning techniques for recognizing and extracting the red light areas arising from the cutting loop in the TURP surgery. First, the ResNet-50 model was used to recognize the red light pattern appearing in the chipped frames of the surgery videos. Then, the proposed Res-Unet model was used to segment the areas with the red light pattern and remove these areas. Finally, the hue, saturation, and value color space were used to classify the four levels of the blood loss under the circumstances of non-red light pattern images. The experiments have shown that the proposed Res-Unet model achieves higher accuracy than other segmentation algorithms in classifying the images with the red and non-red lights, and is able to extract the red light patterns and effectively remove them in the TURP surgery images. The proposed approaches presented here are capable of obtaining the level classifications of blood loss, which are helpful for physicians in diagnosis.

Novel and efficient tag SNPs selection algorithms

SNPs are the most abundant forms of genetic variations amongst species; the association studies between complex diseases and SNPs or haplotypes have received great attention. However, these studies are restricted by the cost of genotyping all SNPs; thus, it is necessary to find smaller subsets, or tag SNPs, representing the rest of the SNPs. In fact, the existing tag SNP selection algorithms are notoriously time-consuming. An efficient algorithm for tag SNP selection was presented, which was applied to analyze the HapMap YRI data. The experimental results show that the proposed algorithm can achieve better performance than the existing tag SNP selection algorithms; in most cases, this proposed algorithm is at least ten times faster than the existing methods. In many cases, when the redundant ratio of the block is high, the proposed algorithm can even be thousands times faster than the previously known methods. Tools and web services for haplotype block analysis integrated by hadoop MapReduce framework are also developed using the proposed algorithm as computation kernels.

NCNet: Deep Learning Network Models for Predicting Function of Non-coding DNA

The human genome consists of 98.5% non-coding DNA sequences, and most of them have no known function. However, a majority of disease-associated variants lie in these regions. Therefore, it is critical to predict the function of non-coding DNA. Hence, we propose the NCNet, which integrates deep residual learning and sequence-to-sequence learning networks, to predict the transcription factor (TF) binding sites, which can then be used to predict non-coding functions. In NCNet, deep residual learning networks are used to enhance the identification rate of regulatory patterns of motifs, so that the sequence-to-sequence learning network may make the most out of the sequential dependency between the patterns. With the identity shortcut technique and deep architectures of the networks, NCNet achieves significant improvement compared to the original hybrid model in identifying regulatory markers.

GPU‐UPGMA: high‐performance computing for UPGMA algorithm based on graphics processing units

Constructing phylogenetic trees is of priority concern in computational biology, especially for developing biological taxonomies. As a conventional means of constructing phylogenetic trees, unweighted pair group method with arithmetic (UPGMA) is also an extensively adopted heuristic algorithm for constructing ultrametric trees (UT). Although the UT constructed by UPGMA is often not a true tree unless the molecular clock assumption holds, UT is still useful for the clocklike data. Moreover, UT has been successfully adopted in other problems, including orthologous‐domain classification and multiple sequence alignment. However, previous implementations of the UPGMA method have a limited ability to handle large taxa sets efficiently. This work describes a novel graphics processing unit (GPU)‐UPGMA approach, capable of providing rapid construction of extremely large datasets for biologists. Experimental results indicate that the proposed GPU‐UPGMA approach achieves an approximately 95× speedup ratio on NVIDIA Tesla C2050 GPU over the implementation with 2.13 GHz CPU. The developed techniques in GPU‐UPGMA also can be applied to solve the classification problem for large data set with more than tens of thousands items in the future.Copyright © 2014 John Wiley & Sons, Ltd.

Efficient haplotype block partitioning and tag SNP selection algorithms under various constraints

Patterns of linkage disequilibrium plays a central role in genome-wide association studies aimed at identifying genetic variation responsible for common human diseases. These patterns in human chromosomes show a block-like structure, and regions of high linkage disequilibrium are called haplotype blocks. A small subset of SNPs, called tag SNPs, is sufficient to capture the haplotype patterns in each haplotype block. Previously developed algorithms completely partition a haplotype sample into blocks while attempting to minimize the number of tag SNPs. However, when resource limitations prevent genotyping all the tag SNPs, it is desirable to restrict their number. We propose two dynamic programming algorithms, incorporating many diversity evaluation functions, for haplotype block partitioning using a limited number of tag SNPs. We use the proposed algorithms to partition the chromosome 21 haplotype data. When the sample is fully partitioned into blocks by our algorithms, the 2,266 blocks and 3,260 tag SNPs are fewer than those identified by previous studies. We also demonstrate that our algorithms find the optimal solution by exploiting the nonmonotonic property of a common haplotype-evaluation function.