Improved image classification explainability with high-accuracy heatmaps.
Ontology highlight
ABSTRACT:
SUBMITTER: Preechakul K
PROVIDER: S-EPMC8889368 | biostudies-literature | 2022 Mar
REPOSITORIES: biostudies-literature
Publications
Improved image classification explainability with high-accuracy heatmaps.
Preechakul Konpat K Sriswasdi Sira S Kijsirikul Boonserm B Chuangsuwanich Ekapol E
iScience 20220215 3
Deep learning models have become increasingly used for image-based classification. In critical applications such as medical imaging, it is important to convey the reasoning behind the models' decisions in human-understandable forms. In this work, we propose Pyramid Localization Network (PYLON), a deep learning model that delivers precise location explanation by increasing the resolution of heatmaps produced by class activation map (CAM). PYLON substantially improves the quality of CAM's heatmaps ...[more]
Similar Datasets
Project description:While convolutional operation effectively extracts local features, their limited receptive fields make it challenging to capture global dependencies. Transformer, on the other hand, excels at global modeling and effectively captures global dependencies. However, the self-attention mechanism used in Transformers lacks a local mechanism for information exchange within specific regions. This article attempts to leverage the strengths of both Transformers and convolutional neural networks (CNNs) to enhance the Swin Transformer V2 model. By incorporating both convolutional operation and self-attention mechanism, the enhanced model combines the local information-capturing capability of CNNs and the long-range dependency-capturing ability of Transformers. The improved model enhances the extraction of local information through the introduction of the Swin Transformer Stem, inverted residual feed-forward network, and Dual-Branch Downsampling structure. Subsequently, it models global dependencies using the improved self-attention mechanism. Additionally, downsampling is applied to the attention mechanism's Q and K to reduce computational and memory overhead. Under identical training conditions, the proposed method significantly improves classification accuracy on multiple image classification datasets, showcasing more robust generalization capabilities.
| S-EPMC10703013 | biostudies-literature
Project description:Small sample learning aims to learn information about object categories from a single or a few training samples. This learning style is crucial for deep learning methods based on large amounts of data. The deep learning method can solve small sample learning through the idea of meta-learning "how to learn by using previous experience." Therefore, this paper takes image classification as the research object to study how meta-learning quickly learns from a small number of sample images. The main contents are as follows: After considering the distribution difference of data sets on the generalization performance of measurement learning and the advantages of optimizing the initial characterization method, this paper adds the model-independent meta-learning algorithm and designs a multi-scale meta-relational network. First, the idea of META-SGD is adopted, and the inner learning rate is taken as the learning vector and model parameter to learn together. Secondly, in the meta-training process, the model-independent meta-learning algorithm is used to find the optimal parameters of the model. The inner gradient iteration is canceled in the process of meta-validation and meta-test. The experimental results show that the multi-scale meta-relational network makes the learned measurement have stronger generalization ability, which further improves the classification accuracy on the benchmark set and avoids the need for fine-tuning of the model-independent meta-learning algorithm.
| S-EPMC8323718 | biostudies-literature
Project description:Automated image-based plant identification has experienced rapid development and has been already used in research and nature management. However, there is a need for extensive studies on how accurately automatic plant identification works and which characteristics of observations and study species influence the results. We investigated the accuracy of the Flora Incognita application, a research-based tool for automated plant image identification. Our study was conducted in Estonia, Northern Europe. Photos originated from the Estonian national curated biodiversity observations database, originally without the intention to use them for automated identification (1496 photos, 542 species) were examined. Flora Incognita was also directly tested in field conditions in various habitats, taking images of plant organs as guided by the application (998 observations, 1703 photos, 280 species). Identification accuracy was compared among species characteristics: plant family, growth forms and life forms, habitat type and regional frequency. We also analysed image characteristics (plant organs, background, number of species in focus), and the number of training images that were available for particular species to develop the automated identification algorithm. From database images 79.6 % of species were correctly identified by Flora Incognita; in the field conditions species identification accuracy reached 85.3 %. Overall, the correct genus was found for 89 % and the correct plant family for 95 % of the species. Accuracy varied among different plant families, life forms and growth forms. Rare and common species and species from different habitats were identified with equal accuracy. Images with reproductive organs or with only the target species in focus were identified with greater success. The number of training images per species was positively correlated with the identification success. Even though a high accuracy has been already achieved for Flora Incognita, allowing its usage for research and practices, our results can guide further improvements of this application and automated plant identification in general.
| S-EPMC8387968 | biostudies-literature
Project description:There remains a large discrepancy between the known genetic contributions to cancer and that which can be explained by genomic variants, both inherited and somatic. Recently, understudied repetitive DNA regions called microsatellites have been identified as genetic risk markers for a number of diseases including various cancers (breast, ovarian and brain). In this study, we demonstrate an integrated process for identifying and further evaluating microsatellite-based risk markers for lung cancer using data from the cancer genome atlas and the 1000 genomes project. Comparing whole-exome germline sequencing data from 488 TCGA lung cancer samples to germline exome data from 390 control samples from the 1000 genomes project, we identified 119 potentially informative microsatellite loci. These loci were found to be able to distinguish between cancer and control samples with sensitivity and specificity ratios over 0.8. Then these loci, supplemented with additional loci from other cancers and controls, were evaluated using a target enrichment kit and sample-multiplexed nextgen sequencing. Thirteen of the 119 risk markers were found to be informative in a well powered study (>0.99 for a 0.95 confidence interval) using high-depth (579x±315) nextgen sequencing of 30 lung cancer and 89 control samples, resulting in sensitivity and specificity ratios of 0.90 and 0.94, respectively. When 8 loci harvested from the bioinformatic analysis of other cancers are added to the classifier, then the sensitivity and specificity rise to 0.93 and 0.97, respectively. Analysis of the genes harboring these loci revealed two genes (ARID1B and REL) and two significantly enriched pathways (chromatin organization and cellular stress response) suggesting that the process of lung carcinogenesis is linked to chromatin remodeling, inflammation, and tumor microenvironment restructuring. We illustrate that high-depth sequencing enables a high-precision microsatellite-based risk classifier analysis approach. This microsatellite-based platform confirms the potential to create clinically actionable diagnostics for lung cancer.
| S-EPMC5701090 | biostudies-literature
Project description:Fluorescence microscopy (FM) and electron microscopy (EM) are complementary techniques. FM affords examination of large fields of view and identifying regions of interest but has a low resolution. EM exhibits excellent resolution over a limited field of view. The combination of these two techniques, correlative microscopy, received considerable interest in the past years and has proven its potential in biology and material science. Accurate correlation of FM and EM images is, however, challenging due to the differences in contrast mechanism, size of field of view and resolution. We report an accurate, fast and robust method to correlate FM and EM images using low densities of fiducial markers. Here, 120 nm diameter fiducial markers consisting of fluorescently labelled silica coated gold nanoparticles are used. The method relies on recording FM, low magnification EM and high magnification EM images. Two linear transformation matrices are constructed, FM to low magnification EM and low magnification EM to high magnification EM. Combination of these matrices results in a high accuracy transformation of FM to high magnification EM coordinates. The method was tested using two different transmission electron microscopes and different Tokuyasu and Lowicryl sections. The overall accuracy of the correlation method is high, 5-30 nm.
| S-EPMC6397213 | biostudies-literature
Project description:ObjectiveTo investigate how the general public trades off explainability versus accuracy of artificial intelligence (AI) systems and whether this differs between healthcare and non-healthcare scenarios.Materials and methodsCitizens' juries are a form of deliberative democracy eliciting informed judgment from a representative sample of the general public around policy questions. We organized two 5-day citizens' juries in the UK with 18 jurors each. Jurors considered 3 AI systems with different levels of accuracy and explainability in 2 healthcare and 2 non-healthcare scenarios. Per scenario, jurors voted for their preferred system; votes were analyzed descriptively. Qualitative data on considerations behind their preferences included transcribed audio-recordings of plenary sessions, observational field notes, outputs from small group work and free-text comments accompanying jurors' votes; qualitative data were analyzed thematically by scenario, per and across AI systems.ResultsIn healthcare scenarios, jurors favored accuracy over explainability, whereas in non-healthcare contexts they either valued explainability equally to, or more than, accuracy. Jurors' considerations in favor of accuracy regarded the impact of decisions on individuals and society, and the potential to increase efficiency of services. Reasons for emphasizing explainability included increased opportunities for individuals and society to learn and improve future prospects and enhanced ability for humans to identify and resolve system biases.ConclusionCitizens may value explainability of AI systems in healthcare less than in non-healthcare domains and less than often assumed by professionals, especially when weighed against system accuracy. The public should therefore be actively consulted when developing policy on AI explainability.
| S-EPMC8522832 | biostudies-literature
Project description:Vehicle image recognition is a critical research area with diverse traffic management, surveillance, and autonomous driving systems applications. Accurately classifying and identifying vehicles from images play a crucial role in these domains. This work presents two vehicle image datasets: the vehicle type image dataset version 2 (VTID2) and the vehicle make image dataset (VMID). The VTID2 Dataset comprises 4,356 images of Thailand's five most used vehicle types, which enhances diversity and reduces the risk of overfitting problems. This expanded dataset offers a more extensive and varied collection for robust model training and evaluation. This dataset will be valuable for researchers focusing on vehicle image recognition tasks. With an emphasis on sedans, hatchbacks, pick-ups, SUVs, and other vehicles, the dataset allows for developing and evaluating algorithms that accurately classify different types of vehicles. The VMID Dataset contains 2,072 images of logos (called vehicle make) from eleven prominent vehicle brands in Thailand. The proposed dataset will facilitate the development of computer vision algorithms and the evaluation of learning algorithm model performance metrics. These two datasets provide valuable resources to the research community that will foster possible research advancements in vehicle recognition, vehicle logo detection or localization, and vehicle segmentation, contributing to the development of intelligent transportation systems.
| S-EPMC10859268 | biostudies-literature
Project description:Stem cell-based tests for developmental toxicity are currently intensively studied. Here, we used a protocol that differentiates human induced pluripotent stem cells (hiPSCs) to neural ectodermal progenitor cells (UKN1) and exposed them to 23 teratogens and 16 non-teratogens at human relevant concentrations. Substance-induced, genome-wide expression profiles were analyzed by two procedures that classified the compounds either based on the number of significantly deregulated genes or on a classification technique with cross-validation. Both procedures demonstrated a high performance with accuracy, sensitivity and specificity of 0.90, 0.83 and 1.0, as well as 0.87, 0.92 and 0.80, respectively. Comparing the present findings with the UKN1 test to those from our recently published UKK2 test that instead recapitulates cardiomyogenic differentiation showed a high congruence in compound classification. This may be explained by the fact that the same pathways, such as PI3K-Akt, P53, TGF-beta, MAPK, EGFR and Hippo were overrepresented among the differentially expressed genes for both tests. However, for some teratogens discrepant results were obtained, such as atorvastatin and thalidomide. In conclusion, UKN1 identifies teratogens with high sensitivity and specificity, and combined interpretation analysis with other tests that differentiate hiPSCs into otherdifferent cell types may further improve the results in the future.
2022-11-03 | GSE209962 | GEO
Project description:Scientists are increasingly using volunteer efforts of citizen scientists to classify images captured by motion-activated trail cameras. The rising popularity of citizen science reflects its potential to engage the public in conservation science and accelerate processing of the large volume of images generated by trail cameras. While image classification accuracy by citizen scientists can vary across species, the influence of other factors on accuracy is poorly understood. Inaccuracy diminishes the value of citizen science derived data and prompts the need for specific best-practice protocols to decrease error. We compare the accuracy between three programs that use crowdsourced citizen scientists to process images online: Snapshot Serengeti, Wildwatch Kenya, and AmazonCam Tambopata. We hypothesized that habitat type and camera settings would influence accuracy. To evaluate these factors, each photograph was circulated to multiple volunteers. All volunteer classifications were aggregated to a single best answer for each photograph using a plurality algorithm. Subsequently, a subset of these images underwent expert review and were compared to the citizen scientist results. Classification errors were categorized by the nature of the error (e.g., false species or false empty), and reason for the false classification (e.g., misidentification). Our results show that Snapshot Serengeti had the highest accuracy (97.9%), followed by AmazonCam Tambopata (93.5%), then Wildwatch Kenya (83.4%). Error type was influenced by habitat, with false empty images more prevalent in open-grassy habitat (27%) compared to woodlands (10%). For medium to large animal surveys across all habitat types, our results suggest that to significantly improve accuracy in crowdsourced projects, researchers should use a trail camera set up protocol with a burst of three consecutive photographs, a short field of view, and determine camera sensitivity settings based on in situ testing. Accuracy level comparisons such as this study can improve reliability of future citizen science projects, and subsequently encourage the increased use of such data.
| S-EPMC7663993 | biostudies-literature
Project description:Progressive supranuclear palsy (PSP), multiple system atrophy (MSA) and idiopathic Parkinson's disease (IPD) can be clinically indistinguishable, especially in the early stages, despite distinct patterns of molecular pathology. Structural neuroimaging holds promise for providing objective biomarkers for discriminating these diseases at the single subject level but all studies to date have reported incomplete separation of disease groups. In this study, we employed multi-class pattern recognition to assess the value of anatomical patterns derived from a widely available structural neuroimaging sequence for automated classification of these disorders. To achieve this, 17 patients with PSP, 14 with IPD and 19 with MSA were scanned using structural MRI along with 19 healthy controls (HCs). An advanced probabilistic pattern recognition approach was employed to evaluate the diagnostic value of several pre-defined anatomical patterns for discriminating the disorders, including: (i) a subcortical motor network; (ii) each of its component regions and (iii) the whole brain. All disease groups could be discriminated simultaneously with high accuracy using the subcortical motor network. The region providing the most accurate predictions overall was the midbrain/brainstem, which discriminated all disease groups from one another and from HCs. The subcortical network also produced more accurate predictions than the whole brain and all of its constituent regions. PSP was accurately predicted from the midbrain/brainstem, cerebellum and all basal ganglia compartments; MSA from the midbrain/brainstem and cerebellum and IPD from the midbrain/brainstem only. This study demonstrates that automated analysis of structural MRI can accurately predict diagnosis in individual patients with Parkinsonian disorders, and identifies distinct patterns of regional atrophy particularly useful for this process.
| S-EPMC3711905 | biostudies-literature