Selective Search for Object Recognition Uijlings, J. R. R.; van de Sande, K. E. A.; Gevers, T. ...
International journal of computer vision,
09/2013, Letnik:
104, Številka:
2
Journal Article
Recenzirano
Odprti dostop
This paper addresses the problem of generating possible object locations for use in object recognition. We introduce selective search which combines the strength of both an exhaustive search and ...segmentation. Like segmentation, we use the image structure to guide our sampling process. Like exhaustive search, we aim to capture all possible object locations. Instead of a single technique to generate possible object locations, we diversify our search and use a variety of complementary image partitionings to deal with as many image conditions as possible. Our selective search results in a small set of data-driven, class-independent, high quality locations, yielding 99 % recall and a Mean Average Best Overlap of 0.879 at 10,097 locations. The reduced number of locations compared to an exhaustive search enables the use of stronger machine learning techniques and stronger appearance models for object recognition. In this paper we show that our selective search enables the use of the powerful Bag-of-Words model for recognition. The selective search software is made publicly available (Software:
http://disi.unitn.it/~uijlings/SelectiveSearch.html
).
Deep learning convolutional neural networks (CNN) may facilitate melanoma detection, but data comparing a CNN’s diagnostic performance to larger groups of dermatologists are lacking.
Google’s ...Inception v4 CNN architecture was trained and validated using dermoscopic images and corresponding diagnoses. In a comparative cross-sectional reader study a 100-image test-set was used (level-I: dermoscopy only; level-II: dermoscopy plus clinical information and images). Main outcome measures were sensitivity, specificity and area under the curve (AUC) of receiver operating characteristics (ROC) for diagnostic classification (dichotomous) of lesions by the CNN versus an international group of 58 dermatologists during level-I or -II of the reader study. Secondary end points included the dermatologists’ diagnostic performance in their management decisions and differences in the diagnostic performance of dermatologists during level-I and -II of the reader study. Additionally, the CNN’s performance was compared with the top-five algorithms of the 2016 International Symposium on Biomedical Imaging (ISBI) challenge.
In level-I dermatologists achieved a mean (±standard deviation) sensitivity and specificity for lesion classification of 86.6% (±9.3%) and 71.3% (±11.2%), respectively. More clinical information (level-II) improved the sensitivity to 88.9% (±9.6%, P=0.19) and specificity to 75.7% (±11.7%, P<0.05). The CNN ROC curve revealed a higher specificity of 82.5% when compared with dermatologists in level-I (71.3%, P<0.01) and level-II (75.7%, P<0.01) at their sensitivities of 86.6% and 88.9%, respectively. The CNN ROC AUC was greater than the mean ROC area of dermatologists (0.86 versus 0.79, P<0.01). The CNN scored results close to the top three algorithms of the ISBI 2016 challenge.
For the first time we compared a CNN’s diagnostic performance with a large international group of 58 dermatologists, including 30 experts. Most dermatologists were outperformed by the CNN. Irrespective of any physicians’ experience, they may benefit from assistance by a CNN’s image classification.
This study was registered at the German Clinical Trial Register (DRKS-Study-ID: DRKS00013570; https://www.drks.de/drks_web/).
Predictive models ground many state-of-the-art developments in statistical brain image analysis: decoding, MVPA, searchlight, or extraction of biomarkers. The principled approach to establish their ...validity and usefulness is cross-validation, testing prediction on unseen data. Here, I would like to raise awareness on error bars of cross-validation, which are often underestimated. Simple experiments show that sample sizes of many neuroimaging studies inherently lead to large error bars, eg±10% for 100 samples. The standard error across folds strongly underestimates them. These large error bars compromise the reliability of conclusions drawn with predictive models, such as biomarkers or methods developments where, unlike with cognitive neuroimaging MVPA approaches, more samples cannot be acquired by repeating the experiment across many subjects. Solutions to increase sample size must be investigated, tackling possible increases in heterogeneity of the data.
UK Biobank is a large-scale prospective epidemiological study with all data accessible to researchers worldwide. It is currently in the process of bringing back 100,000 of the original participants ...for brain, heart and body MRI, carotid ultrasound and low-dose bone/fat x-ray. The brain imaging component covers 6 modalities (T1, T2 FLAIR, susceptibility weighted MRI, Resting fMRI, Task fMRI and Diffusion MRI). Raw and processed data from the first 10,000 imaged subjects has recently been released for general research access. To help convert this data into useful summary information we have developed an automated processing and QC (Quality Control) pipeline that is available for use by other researchers. In this paper we describe the pipeline in detail, following a brief overview of UK Biobank brain imaging and the acquisition protocol. We also describe several quantitative investigations carried out as part of the development of both the imaging protocol and the processing pipeline.
A set of 28 different meteorites was tested in 32 ablation experiments in the plasma wind tunnel PWK1 at the Institute of Space Systems. All meteorites were exposed to the same flow condition in ...consecutive experiments. This paper presents the detailed analysis of high-resolution images taken by DSLR cameras during 26 experiments on 22 different meteorites. It is seen that the ablation behavior of the meteorites differs in the way the material melts and flows downstream. While some meteorites appeared more viscous and most material remained connected to the main body, other samples suggest a much lower viscosity as the material was carried downstream and released droplets to the flow. Most droplets of molten material were seen for the two most carbon rich samples from meteorites Murchison and Dhofar 1575. The release of blue colored particles was observed for several meteorites, independent of the viscosity. In contrast to the molten droplets, the blue particles also traveled up to a few millimeters upstream. The abundance of these blue particles is linked to the iron content. Close to no particles or droplets were observed for achondritic samples.
•Analysis of high-resolution images from meteorite ground experiments.•Ejection of blue colored particles observed with a significant correlation to the iron density.•No particles visible in achondrite experiments.
The amygdala and the hippocampus are two limbic structures that play a critical role in cognition and behavior, however their manual segmentation and that of their smaller nuclei/subfields in ...multicenter datasets is time consuming and difficult due to the low contrast of standard MRI. Here, we assessed the reliability of the automated segmentation of amygdalar nuclei and hippocampal subfields across sites and vendors using FreeSurfer in two independent cohorts of older and younger healthy adults.
Sixty-five healthy older (cohort 1) and 68 younger subjects (cohort 2), from the PharmaCog and CoRR consortia, underwent repeated 3D-T1 MRI (interval 1–90 days). Segmentation was performed using FreeSurfer v6.0. Reliability was assessed using volume reproducibility error (ε) and spatial overlapping coefficient (DICE) between test and retest session.
Significant MRI site and vendor effects (p < .05) were found in a few subfields/nuclei for the ε, while extensive effects were found for the DICE score of most subfields/nuclei. Reliability was strongly influenced by volume, as ε correlated negatively and DICE correlated positively with volume size of structures (absolute value of Spearman’s r correlations >0.43, p < 1.39E-36). In particular, volumes larger than 200 mm3 (for amygdalar nuclei) and 300 mm3 (for hippocampal subfields, except for molecular layer) had the best test-retest reproducibility (ε < 5% and DICE > 0.80).
Our results support the use of volumetric measures of larger amygdalar nuclei and hippocampal subfields in multisite MRI studies. These measures could be useful for disease tracking and assessment of efficacy in drug trials.
•Differences in MRI site/vendor had a limited effect on volume reproducibility.•Differences in MRI site/vendor had an extensive effect on spatial accuracy.•Reliability is good for larger amygdalar and hippocampal structures.•Automated volumetry is reliable in multicenter MRI studies.
Live imaging of large biological specimens is fundamentally limited by the short optical penetration depth of light microscopes. To maximize physical coverage, we developed the SiMView technology ...framework for high-speed in vivo imaging, which records multiple views of the specimen simultaneously. SiMView consists of a light-sheet microscope with four synchronized optical arms, real-time electronics for long-term sCMOS-based image acquisition at 175 million voxels per second, and computational modules for high-throughput image registration, segmentation, tracking and real-time management of the terabytes of multiview data recorded per specimen. We developed one-photon and multiphoton SiMView implementations and recorded cellular dynamics in entire Drosophila melanogaster embryos with 30-s temporal resolution throughout development. We furthermore performed high-resolution long-term imaging of the developing nervous system and followed neuroblast cell lineages in vivo. SiMView data sets provide quantitative morphological information even for fast global processes and enable accurate automated cell tracking in the entire early embryo.
As medical imaging enters its information era and presents rapidly increasing needs for big data analytics, robust pooling and harmonization of imaging data across diverse cohorts with varying ...acquisition protocols have become critical. We describe a comprehensive effort that merges and harmonizes a large-scale dataset of 10,477 structural brain MRI scans from participants without a known neurological or psychiatric disorder from 18 different studies that represent geographic diversity. We use this dataset and multi-atlas-based image processing methods to obtain a hierarchical partition of the brain from larger anatomical regions to individual cortical and deep structures and derive age trends of brain structure through the lifespan (3–96 years old). Critically, we present and validate a methodology for harmonizing this pooled dataset in the presence of nonlinear age trends. We provide a web-based visualization interface to generate and present the resulting age trends, enabling future studies of brain structure to compare their data with this reference of brain development and aging, and to examine deviations from ranges, potentially related to disease.
•Multi-site harmonization method that pools volumetric data from 18 studies, controlling for nonlinear age effects.•Resulting dataset covers ages 3 to 96 and used to derive age trends of brain structure through the lifespan.•Interactive visualization tool provided for exploring age trends and comparing new data.
PDF
