Dataset Information

Harnessing Large-Scale Herbarium Image Datasets Through Representation Learning.

ABSTRACT: The mobilization of large-scale datasets of specimen images and metadata through herbarium digitization provide a rich environment for the application and development of machine learning techniques. However, limited access to computational resources and uneven progress in digitization, especially for small herbaria, still present barriers to the wide adoption of these new technologies. Using deep learning to extract representations of herbarium specimens useful for a wide variety of applications, so-called "representation learning," could help remove these barriers. Despite its recent popularity for camera trap and natural world images, representation learning is not yet as popular for herbarium specimen images. We investigated the potential of representation learning with specimen images by building three neural networks using a publicly available dataset of over 2 million specimen images spanning multiple continents and institutions. We compared the extracted representations and tested their performance in application tasks relevant to research carried out with herbarium specimens. We found a triplet network, a type of neural network that learns distances between images, produced representations that transferred the best across all applications investigated. Our results demonstrate that it is possible to learn representations of specimen images useful in different applications, and we identify some further steps that we believe are necessary for representation learning to harness the rich information held in the worlds' herbaria.

SUBMITTER: Walker BE

PROVIDER: S-EPMC8794728 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Much ecological research relies on existing multispecies distribution datasets. Such datasets, however, can vary considerably in quality, extent, resolution or taxonomic coverage. We provide a framework for a spatially-explicit evaluation of geographical representation within large-scale species distribution datasets, using the comparison of an occurrence atlas with a range atlas dataset as a working example. Specifically, we compared occurrence maps for 3773 taxa from the widely-used Atlas Florae Europaeae (AFE) with digitised range maps for 2049 taxa of the lesser-known Atlas of North European Vascular Plants. We calculated the level of agreement at a 50-km spatial resolution using average latitudinal and longitudinal species range, and area of occupancy. Agreement in species distribution was calculated and mapped using Jaccard similarity index and a reduced major axis (RMA) regression analysis of species richness between the entire atlases (5221 taxa in total) and between co-occurring species (601 taxa). We found no difference in distribution ranges or in the area of occupancy frequency distribution, indicating that atlases were sufficiently overlapping for a valid comparison. The similarity index map showed high levels of agreement for central, western, and northern Europe. The RMA regression confirmed that geographical representation of AFE was low in areas with a sparse data recording history (e.g., Russia, Belarus and the Ukraine). For co-occurring species in south-eastern Europe, however, the Atlas of North European Vascular Plants showed remarkably higher richness estimations. Geographical representation of atlas data can be much more heterogeneous than often assumed. Level of agreement between datasets can be used to evaluate geographical representation within datasets. Merging atlases into a single dataset is worthwhile in spite of methodological differences, and helps to fill gaps in our knowledge of species distribution ranges. Species distribution dataset mergers, such as the one exemplified here, can serve as a baseline towards comprehensive species distribution datasets.

Project description:BackgroundBody composition (BC) is an important factor in determining the risk of type 2-diabetes and cardiovascular disease. Computed tomography (CT) is a useful imaging technique for studying BC, however manual segmentation of CT images is time-consuming and subjective. The purpose of this study is to develop and evaluate fully automated segmentation techniques applicable to a 3-slice CT imaging protocol, consisting of single slices at the level of the liver, abdomen, and thigh, allowing detailed analysis of numerous tissues and organs.MethodsThe study used more than 4000 CT subjects acquired from the large-scale SCAPIS and IGT cohort to train and evaluate four convolutional neural network based architectures: ResUNET, UNET++, Ghost-UNET, and the proposed Ghost-UNET++. The segmentation techniques were developed and evaluated for automated segmentation of the liver, spleen, skeletal muscle, bone marrow, cortical bone, and various adipose tissue depots, including visceral (VAT), intraperitoneal (IPAT), retroperitoneal (RPAT), subcutaneous (SAT), deep (DSAT), and superficial SAT (SSAT), as well as intermuscular adipose tissue (IMAT). The models were trained and validated for each target using tenfold cross-validation and test sets.ResultsThe Dice scores on cross validation in SCAPIS were: ResUNET 0.964 (0.909-0.996), UNET++ 0.981 (0.927-0.996), Ghost-UNET 0.961 (0.904-0.991), and Ghost-UNET++ 0.968 (0.910-0.994). All four models showed relatively strong results, however UNET++ had the best performance overall. Ghost-UNET++ performed competitively compared to UNET++ and showed a more computationally efficient approach.ConclusionFully automated segmentation techniques can be successfully applied to a 3-slice CT imaging protocol to analyze multiple tissues and organs related to BC. The overall best performance was achieved by UNET++, against which Ghost-UNET++ showed competitive results based on a more computationally efficient approach. The use of fully automated segmentation methods can reduce analysis time and provide objective results in large-scale studies of BC.

Project description:PurposeThe curation of images using human resources is time intensive but an essential step for developing artificial intelligence (AI) algorithms. Our goal was to develop and implement an AI algorithm for image curation in a high-volume setting. We also explored AI tools that will assist in deploying a tiered approach, in which the AI model labels images and flags potential mislabels for human review.DesignImplementation of an AI algorithm.ParticipantsSeven-field stereoscopic images from multiple clinical trials.MethodsThe 7-field stereoscopic image protocol includes 7 pairs of images from various parts of the central retina along with images of the anterior part of the eye. All images were labeled for field number by reading center graders. The model output included classification of the retinal images into 8 field numbers. Probability scores (0-1) were generated to identify misclassified images, with 1 indicating a high probability of a correct label.Main outcome measuresAgreement of AI prediction with grader classification of field number and the use of probability scores to identify mislabeled images.ResultsThe AI model was trained and validated on 17 529 images and tested on 3004 images. The pooled agreement of field numbers between grader classification and the AI model was 88.3% (kappa, 0.87). The pooled mean probability score was 0.97 (standard deviation [SD], 0.08) for images for which the graders agreed with the AI-generated labels and 0.77 (SD, 0.19) for images for which the graders disagreed with the AI-generated labels (P < 0.0001). Using receiver operating characteristic curves, a probability score of 0.99 was identified as a cutoff for distinguishing mislabeled images. A tiered workflow using a probability score of < 0.99 as a cutoff would include 27.6% of the 3004 images for human review and reduce the error rate from 11.7% to 1.5%.ConclusionsThe implementation of AI algorithms requires measures in addition to model validation. Tools to flag potential errors in the labels generated by AI models will reduce inaccuracies, increase trust in the system, and provide data for continuous model development.

Dataset Information

Harnessing Large-Scale Herbarium Image Datasets Through Representation Learning.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets