A NEW STEREO DENSE MATCHING BENCHMARK DATASET FOR DEEP LEARNING

Abstract. Stereo dense matching is a fundamental task for 3D scene reconstruction. Recently, deep learning based methods have proven effective on some benchmark datasets, for example Middlebury and KITTI stereo. However, it is not easy to find a training dataset for aerial photogrammetry. Generating ground truth data for real scenes is a challenging task. In the photogrammetry community, many evaluation methods use digital surface models (DSM) to generate the ground truth disparity for the stereo pairs, but in this case interpolation may bring errors in the estimated disparity. In this paper, we publish a stereo dense matching dataset based on ISPRS Vaihingen dataset, and use it to evaluate some traditional and deep learning based methods. The evaluation shows that learning-based methods outperform traditional methods significantly when the fine tuning is done on a similar landscape. The benchmark also investigates the impact of the base to height ratio on the performance of the evaluated methods. The dataset can be found in https://github.com/whuwuteng/benchmark_ISPRS2021.

Download Full-text

Abstract 20: Deep Learning-Based Automated Intracranial Hemorrhage Detection and Notification

Stroke ◽

10.1161/str.51.suppl_1.20 ◽

2020 ◽

Vol 51 (Suppl_1) ◽

Author(s):

Benjamin Zahneisen ◽

Matus Straka ◽

Shalini Bammer ◽

Greg Albers ◽

Roland Bammer

Keyword(s):

Deep Learning ◽

Sensitivity And Specificity ◽

Intracranial Hemorrhage ◽

Health Care Professionals ◽

Gold Standard ◽

Ground Truth ◽

Mobile App ◽

Training Dataset ◽

Convolutional Network ◽

Ground Truth Data

Introduction: Ruling out hemorrhage (stroke or traumatic) prior to administration of thrombolytics is critical for Code Strokes. A triage software that identifies hemorrhages on head CTs and alerts radiologists would help to streamline patient care and increase diagnostic confidence and patient safety. ML approach: We trained a deep convolutional network with a hybrid 3D/2D architecture on unenhanced head CTs of 805 patients. Our training dataset comprised 348 positive hemorrhage cases (IPH=245, SAH=67, Sub/Epi-dural=70, IVH=83) (128 female) and 457 normal controls (217 female). Lesion outlines were drawn by experts and stored as binary masks that were used as ground truth data during the training phase (random 80/20 train/test split). Diagnostic sensitivity and specificity were defined on a per patient study level, i.e. a single, binary decision for presence/absence of a hemorrhage on a patient’s CT scan. Final validation was performed in 380 patients (167 positive). Tool: The hemorrhage detection module was prototyped in Python/Keras. It runs on a local LINUX server (4 CPUs, no GPUs) and is embedded in a larger image processing platform dedicated to stroke. Results: Processing time for a standard whole brain CT study (3-5mm slices) was around 2min. Upon completion, an instant notification (by email and/or mobile app) was sent to users to alert them about the suspected presence of a hemorrhage. Relative to neuroradiologist gold standard reads the algorithm’s sensitivity and specificity is 90.4% and 92.5% (95% CI: 85%-94% for both). Detection of acute intracranial hemorrhage can be automatized by deploying deep learning. It yielded very high sensitivity/specificity when compared to gold standard reads by a neuroradiologist. Volumes as small as 0.5mL could be detected reliably in the test dataset. The software can be deployed in busy practices to prioritize worklists and alert health care professionals to speed up therapeutic decision processes and interventions.

Download Full-text

EVICAN—a balanced dataset for algorithm development in cell and nucleus segmentation

Bioinformatics ◽

10.1093/bioinformatics/btaa225 ◽

2020 ◽

Vol 36 (12) ◽

pp. 3863-3870

Author(s):

Mischa Schwendy ◽

Ronald E Unger ◽

Sapun H Parekh

Keyword(s):

Deep Learning ◽

Cell Biology ◽

Ground Truth ◽

Training Data ◽

Training Dataset ◽

Visual Cell ◽

Application Development ◽

Ground Truth Data ◽

Quantitative Image ◽

Nucleus Segmentation

Abstract Motivation Deep learning use for quantitative image analysis is exponentially increasing. However, training accurate, widely deployable deep learning algorithms requires a plethora of annotated (ground truth) data. Image collections must contain not only thousands of images to provide sufficient example objects (i.e. cells), but also contain an adequate degree of image heterogeneity. Results We present a new dataset, EVICAN—Expert visual cell annotation, comprising partially annotated grayscale images of 30 different cell lines from multiple microscopes, contrast mechanisms and magnifications that is readily usable as training data for computer vision applications. With 4600 images and ∼26 000 segmented cells, our collection offers an unparalleled heterogeneous training dataset for cell biology deep learning application development. Availability and implementation The dataset is freely available (https://edmond.mpdl.mpg.de/imeji/collection/l45s16atmi6Aa4sI?q=). Using a Mask R-CNN implementation, we demonstrate automated segmentation of cells and nuclei from brightfield images with a mean average precision of 61.6 % at a Jaccard Index above 0.5.

Download Full-text

Deep Learning Based Cardiac MRI Segmentation: Do We Need Experts?

Algorithms ◽

10.3390/a14070212 ◽

2021 ◽

Vol 14 (7) ◽

pp. 212

Author(s):

Youssef Skandarani ◽

Pierre-Marc Jodoin ◽

Alain Lalande

Keyword(s):

Deep Learning ◽

Cardiac Mri ◽

Expert Knowledge ◽

Medical Image Analysis ◽

Ground Truth ◽

Cine Mri ◽

Data Sets ◽

Mri Segmentation ◽

Segmentation Evaluation ◽

Ground Truth Data

Deep learning methods are the de facto solutions to a multitude of medical image analysis tasks. Cardiac MRI segmentation is one such application, which, like many others, requires a large number of annotated data so that a trained network can generalize well. Unfortunately, the process of having a large number of manually curated images by medical experts is both slow and utterly expensive. In this paper, we set out to explore whether expert knowledge is a strict requirement for the creation of annotated data sets on which machine learning can successfully be trained. To do so, we gauged the performance of three segmentation models, namely U-Net, Attention U-Net, and ENet, trained with different loss functions on expert and non-expert ground truth for cardiac cine–MRI segmentation. Evaluation was done with classic segmentation metrics (Dice index and Hausdorff distance) as well as clinical measurements, such as the ventricular ejection fractions and the myocardial mass. The results reveal that generalization performances of a segmentation neural network trained on non-expert ground truth data is, to all practical purposes, as good as that trained on expert ground truth data, particularly when the non-expert receives a decent level of training, highlighting an opportunity for the efficient and cost-effective creation of annotations for cardiac data sets.

Download Full-text

A flexible ChIP-sequencing simulation toolkit

BMC Bioinformatics ◽

10.1186/s12859-021-04097-5 ◽

2021 ◽

Vol 22 (1) ◽

Author(s):

An Zheng ◽

Michael Lamkin ◽

Yutong Qiu ◽

Kevin Ren ◽

Alon Goren ◽

...

Keyword(s):

Ground Truth ◽

Peak Calling ◽

Simulation Framework ◽

Experimental Conditions ◽

Ground Truth Data ◽

Chip Sequencing ◽

Genome Wide ◽

Experimental Parameters ◽

Differential Binding ◽

The Impact

Abstract Background A major challenge in evaluating quantitative ChIP-seq analyses, such as peak calling and differential binding, is a lack of reliable ground truth data. Accurate simulation of ChIP-seq data can mitigate this challenge, but existing frameworks are either too cumbersome to apply genome-wide or unable to model a number of important experimental conditions in ChIP-seq. Results We present ChIPs, a toolkit for rapidly simulating ChIP-seq data using statistical models of key experimental steps. We demonstrate how ChIPs can be used for a range of applications, including benchmarking analysis tools and evaluating the impact of various experimental parameters. ChIPs is implemented as a standalone command-line program written in C++ and is available from https://github.com/gymreklab/chips. Conclusions ChIPs is an efficient ChIP-seq simulation framework that generates realistic datasets over a flexible range of experimental conditions. It can serve as an important component in various ChIP-seq analyses where ground truth data are needed.

Download Full-text

Covid-19 detection via deep neural network and occlusion sensitivity maps

10.36227/techrxiv.14100890 ◽

2021 ◽

Author(s):

Noor Ahmad ◽

Muhammad Aminu ◽

Mohd Halim Mohd Noor

Keyword(s):

Neural Network ◽

Deep Learning ◽

Deep Neural Network ◽

State Of The Art ◽

Color Images ◽

Fine Tuning ◽

Training Dataset ◽

Learning Approaches ◽

Learning Models ◽

Sensitivity Maps

Deep learning approaches have attracted a lot of attention in the automatic detection of Covid-19 and transfer learning is the most common approach. However, majority of the pre-trained models are trained on color images, which can cause inefficiencies when fine-tuning the models on Covid-19 images which are often grayscale. To address this issue, we propose a deep learning architecture called CovidNet which requires a relatively smaller number of parameters. CovidNet accepts grayscale images as inputs and is suitable for training with limited training dataset. Experimental results show that CovidNet outperforms other state-of-the-art deep learning models for Covid-19 detection.

Download Full-text

Encoding Health Records into Pathway Representations for Deep Learning

10.3233/shti210800 ◽

2021 ◽

Author(s):

Marco Luca Sbodio ◽

Natasha Mulligan ◽

Stefanie Speichert ◽

Vanessa Lopez ◽

Joao Bettencourt-Silva

Keyword(s):

Neural Network ◽

Machine Learning ◽

Deep Learning ◽

Source Code ◽

Training Dataset ◽

Health Records ◽

Learning Tasks ◽

Patient Pathways ◽

Computational Resources ◽

The Impact

There is a growing trend in building deep learning patient representations from health records to obtain a comprehensive view of a patient’s data for machine learning tasks. This paper proposes a reproducible approach to generate patient pathways from health records and to transform them into a machine-processable image-like structure useful for deep learning tasks. Based on this approach, we generated over a million pathways from FAIR synthetic health records and used them to train a convolutional neural network. Our initial experiments show the accuracy of the CNN on a prediction task is comparable or better than other autoencoders trained on the same data, while requiring significantly less computational resources for training. We also assess the impact of the size of the training dataset on autoencoders performances. The source code for generating pathways from health records is provided as open source.

Download Full-text

Abstract P319: Can Deep Learning Find the Ischemic Core on CT? Transfer Learning From Pre-Trained MRI-Based Networks

Stroke ◽

10.1161/str.52.suppl_1.p319 ◽

2021 ◽

Vol 52 (Suppl_1) ◽

Author(s):

Yannan Yu ◽

Soren Christensen ◽

Yuan Xie ◽

Enhao Gong ◽

Maarten G Lansberg ◽

...

Keyword(s):

Deep Learning ◽

Ground Truth ◽

Learning Model ◽

Fine Tuning ◽

Learning Models ◽

Starting Point ◽

Stroke Lesion ◽

Ischemic Core ◽

Deep Learning Model

Objective: Ischemic core prediction from CT perfusion (CTP) remains inaccurate compared with gold standard diffusion-weighted imaging (DWI). We evaluated if a deep learning model to predict the DWI lesion from MR perfusion (MRP) could facilitate ischemic core prediction on CTP. Method: Using the multi-center CRISP cohort of acute ischemic stroke patient with CTP before thrombectomy, we included patients with major reperfusion (TICI score≥2b), adequate image quality, and follow-up MRI at 3-7 days. Perfusion parameters including Tmax, mean transient time, cerebral blood flow (CBF), and cerebral blood volume were reconstructed by RAPID software. Core lab experts outlined the stroke lesion on the follow-up MRI. A previously trained MRI model in a separate group of patients was used as a starting point, which used MRP parameters as input and RAPID ischemic core on DWI as ground truth. We fine-tuned this model, using CTP parameters as input, and follow-up MRI as ground truth. Another model was also trained from scratch with only CTP data. 5-fold cross validation was used. Performance of the models was compared with ischemic core (rCBF≤30%) from RAPID software to identify the presence of a large infarct (volume>70 or >100ml). Results: 94 patients in the CRISP trial met the inclusion criteria (mean age 67±15 years, 52% male, median baseline NIHSS 18, median 90-day mRS 2). Without fine-tuning, the MRI model had an agreement of 73% in infarct >70ml, and 69% in >100ml; the MRI model fine-tuned on CT improved the agreement to 77% and 73%; The CT model trained from scratch had agreements of 73% and 71%; All of the deep learning models outperformed the rCBF segmentation from RAPID, which had agreements of 51% and 64%. See Table and figure. Conclusions: It is feasible to apply MRP-based deep learning model to CT. Fine-tuning with CTP data further improves the predictions. All deep learning models predict the stroke lesion after major recanalization better than thresholding approaches based on rCBF.

Download Full-text

Learning Student Networks with Few Data

Proceedings of the AAAI Conference on Artificial Intelligence ◽

10.1609/aaai.v34i04.5874 ◽

2020 ◽

Vol 34 (04) ◽

pp. 4469-4476 ◽

Cited By ~ 1

Author(s):

Shumin Kong ◽

Tianyu Guo ◽

Shan You ◽

Chang Xu

Keyword(s):

Data Privacy ◽

Empirical Distribution ◽

Ground Truth ◽

Wasserstein Distance ◽

Worst Case ◽

Ground Truth Data ◽

Teacher Student ◽

Benchmark Datasets ◽

Training Examples ◽

Few Data

Recently, the teacher-student learning paradigm has drawn much attention in compressing neural networks on low-end edge devices, such as mobile phones and wearable watches. Current algorithms mainly assume the complete dataset for the teacher network is also available for the training of the student network. However, for real-world scenarios, users may only have access to part of training examples due to commercial profits or data privacy, and severe over-fitting issues would happen as a result. In this paper, we tackle the challenge of learning student networks with few data by investigating the ground-truth data-generating distribution underlying these few data. Taking Wasserstein distance as the measurement, we assume this ideal data distribution lies in a neighborhood of the discrete empirical distribution induced by the training examples. Thus we propose to safely optimize the worst-case cost within this neighborhood to boost the generalization. Furthermore, with theoretical analysis, we derive a novel and easy-to-implement loss for training the student network in an end-to-end fashion. Experimental results on benchmark datasets validate the effectiveness of our proposed method.

Download Full-text

Covid-19 detection via deep neural network and occlusion sensitivity maps

10.36227/techrxiv.14100890.v1 ◽

2021 ◽

Author(s):

Noor Ahmad ◽

Muhammad Aminu ◽

Mohd Halim Mohd Noor

Keyword(s):

Neural Network ◽

Deep Learning ◽

Deep Neural Network ◽

State Of The Art ◽

Color Images ◽

Fine Tuning ◽

Training Dataset ◽

Learning Approaches ◽

Learning Models ◽

Sensitivity Maps

Download Full-text

DEEP LEARNING BASED ROOF TYPE CLASSIFICATION USING VERY HIGH RESOLUTION AERIAL IMAGERY

ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences ◽

10.5194/isprs-archives-xliii-b3-2021-55-2021 ◽

2021 ◽

Vol XLIII-B3-2021 ◽

pp. 55-60

Author(s):

M. Buyukdemircioglu ◽

R. Can ◽

S. Kocaman

Keyword(s):

Deep Learning ◽

High Resolution ◽

Urban Areas ◽

Image Features ◽

Training Data ◽

Fine Tuning ◽

Computer Hardware ◽

Geographical Information ◽

Training Dataset ◽

Very High

Abstract. Automatic detection, segmentation and reconstruction of buildings in urban areas from Earth Observation (EO) data are still challenging for many researchers. Roof is one of the most important element in a building model. The three-dimensional geographical information system (3D GIS) applications generally require the roof type and roof geometry for performing various analyses on the models, such as energy efficiency. The conventional segmentation and classification methods are often based on features like corners, edges and line segments. In parallel to the developments in computer hardware and artificial intelligence (AI) methods including deep learning (DL), image features can be extracted automatically. As a DL technique, convolutional neural networks (CNNs) can also be used for image classification tasks, but require large amount of high quality training data for obtaining accurate results. The main aim of this study was to generate a roof type dataset from very high-resolution (10 cm) orthophotos of Cesme, Turkey, and to classify the roof types using a shallow CNN architecture. The training dataset consists 10,000 roof images and their labels. Six roof type classes such as flat, hip, half-hip, gable, pyramid and complex roofs were used for the classification in the study area. The prediction performance of the shallow CNN model used here was compared with the results obtained from the fine-tuning of three well-known pre-trained networks, i.e. VGG-16, EfficientNetB4, ResNet-50. The results show that although our CNN has slightly lower performance expressed with the overall accuracy, it is still acceptable for many applications using sparse data.

Download Full-text