Convolutional neural networks trained with a developmental sequence of blurry to clear images reveal core differences between face and object processing.
Ontology highlight
ABSTRACT:
SUBMITTER: Jang H
PROVIDER: S-EPMC8590164 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:We present two new fisheye image datasets for training object and face detection models: VOC-360 and Wider-360. The fisheye images are created by post-processing regular images collected from two well-known datasets, VOC2012 and Wider Face, using a model for mapping regular to fisheye images implemented in Matlab. VOC-360 contains 39,575 fisheye images for object detection, segmentation, and classification. Wider-360 contains 63,897 fisheye images for face detection. These datasets will be useful for developing face and object detectors as well as segmentation modules for fisheye images while the efforts to collect and manually annotate true fisheye images are underway.
| S-EPMC6920438 | biostudies-literature
Project description:Prosopagnosia has largely been regarded as an untreatable disorder. However, recent case studies using cognitive training have shown that it is possible to enhance face recognition abilities in individuals with developmental prosopagnosia. Our goal was to determine if this approach could be effective in a larger population of developmental prosopagnosics. We trained 24 developmental prosopagnosics using a 3-week online face-training program targeting holistic face processing. Twelve subjects with developmental prosopagnosia were assessed before and after training, and the other 12 were assessed before and after a waiting period, they then performed the training, and were then assessed again. The assessments included measures of front-view face discrimination, face discrimination with view-point changes, measures of holistic face processing, and a 5-day diary to quantify potential real-world improvements. Compared with the waiting period, developmental prosopagnosics showed moderate but significant overall training-related improvements on measures of front-view face discrimination. Those who reached the more difficult levels of training ('better' trainees) showed the strongest improvements in front-view face discrimination and showed significantly increased holistic face processing to the point of being similar to that of unimpaired control subjects. Despite challenges in characterizing developmental prosopagnosics' everyday face recognition and potential biases in self-report, results also showed modest but consistent self-reported diary improvements. In summary, we demonstrate that by using cognitive training that targets holistic processing, it is possible to enhance face perception across a group of developmental prosopagnosics and further suggest that those who improved the most on the training task received the greatest benefits.
| S-EPMC4032098 | biostudies-literature
Project description:Expertise in processing faces is a cornerstone of human social interaction. However, the developmental course of many key brain regions supporting face preferential processing in the human brain remains undefined. Here, we present findings from an FMRI study using a simple viewing paradigm of faces and objects in a continuous age sample covering the age range from 6 years through adulthood. These findings are the first to use such a sample paired with whole-brain FMRI analyses to investigate development within the core and extended face networks across the developmental spectrum from middle childhood to adulthood. We found evidence, albeit modest, for a developmental trend in the volume of the right fusiform face area (rFFA) but no developmental change in the intensity of activation. From a spatial perspective, the middle portion of the right fusiform gyrus most commonly found in adult studies of face processing was increasingly likely to be included in the FFA as age increased to adulthood. Outside of the FFA, the most striking finding was that children hyperactivated nearly every aspect of the extended face system relative to adults, including the amygdala, anterior temporal pole, insula, inferior frontal gyrus, anterior cingulate gyrus, and parietal cortex. Overall, the findings suggest that development is best characterized by increasing modulation of face-sensitive regions throughout the brain to engage only those systems necessary for task requirements.
| S-EPMC3825803 | biostudies-literature
Project description:Deep convolutional neural networks (DCNNs) trained for face identification can rival and even exceed human-level performance. The ways in which the internal face representations in DCNNs relate to human cognitive representations and brain activity are not well understood. Nearly all previous studies focused on static face image processing with rapid display times and ignored the processing of naturalistic, dynamic information. To address this gap, we developed the largest naturalistic dynamic face stimulus set in human neuroimaging research (700+ naturalistic video clips of unfamiliar faces). We used this naturalistic dataset to compare representational geometries estimated from DCNNs, behavioral responses, and brain responses. We found that DCNN representational geometries were consistent across architectures, cognitive representational geometries were consistent across raters in a behavioral arrangement task, and neural representational geometries in face areas were consistent across brains. Representational geometries in late, fully connected DCNN layers, which are optimized for individuation, were much more weakly correlated with cognitive and neural geometries than were geometries in late-intermediate layers. The late-intermediate face-DCNN layers successfully matched cognitive representational geometries, as measured with a behavioral arrangement task that primarily reflected categorical attributes, and correlated with neural representational geometries in known face-selective topographies. Our study suggests that current DCNNs successfully capture neural cognitive processes for categorical attributes of faces but less accurately capture individuation and dynamic features.
| S-EPMC10614847 | biostudies-literature
Project description:BackgroundFace perception impairment in schizophrenia has long been recognized. However, brain mechanisms underlying this socially important perceptual deficit are not well understood. Previous magnetic resonance imaging (MRI) studies have shown that patients have altered structure in brain regions responsible for processing face information, but functional properties of these brain regions are not clearly determined. A key functional property of the face-processing system--face selectivity--has yet to be evaluated in schizophrenia.MethodsWe used functional MRI (fMRI) to examine face selectivity of 3 core face-processing regions--fusiform face area (FFA), occipital face area (OFA), and superior temporal sulcus (STS)--in schizophrenia patients (n = 24) and healthy controls (n = 23). To disassociate face-specific processing from general perceptual processing, we compared cortical activations during performance of perceptually equated face and tree detection tasks.ResultsActivation levels of the 3 putative face-processing regions during face detection did not differ between patients and controls, being similar for FFA and OFA and absent for STS. However, face selectivity, indexed by the difference in cortical activation between face and tree detection, was significantly reduced in patients for FFA, especially for low-contrast stimuli. FFA activation and perceptual performance during face detection were associated in patients.ConclusionsThese results show a lack of face-specific processing in the schizophrenic brain region presumably subserving face perception. This finding suggests boosting visual salience of face images as a potential therapeutic venue for improving face perception in this psychiatric disorder.
| S-EPMC4838078 | biostudies-literature
Project description:Histopathological images contain rich phenotypic descriptions of the molecular processes underlying disease progression. Convolutional neural networks, state-of-the-art image analysis techniques in computer vision, automatically learn representative features from such images which can be useful for disease diagnosis, prognosis, and subtyping. Hepatocellular carcinoma (HCC) is the sixth most common type of primary liver malignancy. Despite the high mortality rate of HCC, little previous work has made use of CNN models to explore the use of histopathological images for prognosis and clinical survival prediction of HCC. We applied three pre-trained CNN models-VGG 16, Inception V3 and ResNet 50-to extract features from HCC histopathological images. Sample visualization and classification analyses based on these features showed a very clear separation between cancer and normal samples. In a univariate Cox regression analysis, 21.4% and 16% of image features on average were significantly associated with overall survival (OS) and disease-free survival (DFS), respectively. We also observed significant correlations between these features and integrated biological pathways derived from gene expression and copy number variation. Using an elastic net regularized Cox Proportional Hazards model of OS constructed from Inception image features, we obtained a concordance index (C-index) of 0.789 and a significant log-rank test (p = 7.6E-18). We also performed unsupervised classification to identify HCC subgroups from image features. The optimal two subgroups discovered using Inception model image features showed significant differences in both overall (C-index = 0.628 and p = 7.39E-07) and DFS (C-index = 0.558 and p = 0.012). Our work demonstrates the utility of extracting image features using pre-trained models by using them to build accurate prognostic models of HCC as well as highlight significant correlations between these features, clinical survival, and relevant biological pathways. Image features extracted from HCC histopathological images using the pre-trained CNN models VGG 16, Inception V3 and ResNet 50 can accurately distinguish normal and cancer samples. Furthermore, these image features are significantly correlated with survival and relevant biological pathways.
| S-EPMC7073245 | biostudies-literature
Project description:BackgroundImmunofluorescent confocal microscopy uses labeled antibodies as probes against specific macromolecules to discriminate between multiple cell types. For images of the developmental mouse lung, these cells are themselves organized into densely packed higher-level anatomical structures. These types of images can be challenging to segment automatically for several reasons, including the relevance of biomedical context, dependence on the specific set of probes used, prohibitive cost of generating labeled training data, as well as the complexity and dense packing of anatomical structures in the image. The use of an application ontology helps surmount these challenges by combining image data with its metadata to provide a meaningful biological context, modeled after how a human expert would make use of contextual information to identify histological structures, that constrains and simplifies the process of segmentation and object identification.ResultsWe propose an innovative approach for the semi-supervised analysis of complex and densely packed anatomical structures from immunofluorescent images that utilizes an application ontology to provide a simplified context for image segmentation and object identification. We describe how the logical organization of biological facts in the form of an ontology can provide useful constraints that facilitate automatic processing of complex images. We demonstrate the results of ontology-guided segmentation and object identification in mouse developmental lung images from the Bioinformatics REsource ATlas for the Healthy lung database of the Molecular Atlas of Lung Development (LungMAP1) program CONCLUSION: We describe a novel ontology-guided approach to segmentation and classification of complex immunofluorescence images of the developing mouse lung. The ontology is used to automatically generate constraints for each image based on its biomedical context, which facilitates image segmentation and classification.
| S-EPMC7901098 | biostudies-literature
Project description:PurposeArtificial intelligence-enabled techniques can process large amounts of surgical data and may be utilized for clinical decision support to recognize or forecast adverse events in an actual intraoperative scenario. To develop an image-guided navigation technology that will help in surgical education, we explored the performance of a convolutional neural network (CNN)-based computer vision system in detecting intraoperative objects.MethodsThe surgical videos used for annotation were recorded during surgeries conducted in the Department of Surgery of Tokyo Women's Medical University from 2019 to 2020. Abdominal endoscopic images were cut out from manually captured surgical videos. An open-source programming framework for CNN was used to design a model that could recognize and segment objects in real time through IBM Visual Insights. The model was used to detect the GI tract, blood, vessels, uterus, forceps, ports, gauze and clips in the surgical images.ResultsThe accuracy, precision and recall of the model were 83%, 80% and 92%, respectively. The mean average precision (mAP), the calculated mean of the precision for each object, was 91%. Among surgical tools, the highest recall and precision of 96.3% and 97.9%, respectively, were achieved for forceps. Among the anatomical structures, the highest recall and precision of 92.9% and 91.3%, respectively, were achieved for the GI tract.ConclusionThe proposed model could detect objects in operative images with high accuracy, highlighting the possibility of using AI-based object recognition techniques for intraoperative navigation. Real-time object recognition will play a major role in navigation surgery and surgical education.
| S-EPMC8224261 | biostudies-literature
Project description:Rationale: Face expertise is a pivotal social skill. Developmental prosopagnosia (DP), i.e., the inability to recognize faces without a history of brain damage, affects about 2% of the general population, and is a renowned model system of the face-processing network. Within this network, the right Fusiform Face Area (FFA), is particularly involved in face identity processing and may therefore be a key element in DP. Neural representations within the FFA have been examined with Representational Similarity Analysis (RSA), a data-analytical framework in which multi-unit measures of brain activity are assessed with correlation analysis. Objectives: Our study intended to scrutinize modifications of FFA-activation during face encoding and maintenance based on RSA. Methods: Thirteen participants with DP (23-70 years) and 12 healthy control subjects (19-62 years) participated in a functional MRI study, including morphological MRI, a functional FFA-localizer and a modified Sternberg paradigm probing face memory encoding and maintenance. Memory maintenance of one, two, or four faces represented low, medium, and high memory load. We examined conventional activation differences in response to working memory load and applied RSA to compute individual correlation-matrices on the voxel level. Group correlation-matrices were compared via Donsker's random walk analysis. Results: On the functional level, increased memory load entailed both a higher absolute FFA-activation level and a higher degree of correlation between activated voxels. Both aspects were deficient in DP. Interestingly, control participants showed a homogeneous degree of correlation for successful trials during the experiment. In DP-participants, correlation levels between FFA-voxels were significantly lower and were less sustained during the experiment. In behavioral terms, DP-participants performed poorer and had longer reaction times in relation to DP-severity. Furthermore, correlation levels were negatively correlated with reaction times for the most demanding high load condition. Conclusion: We suggest that participants with DP fail to generate robust and maintained neural representations in the FFA during face encoding and maintenance, in line with poorer task performance and prolonged reaction times. In DP, alterations of neural coding in the FFA might therefore explain curtailing in working memory and contribute to impaired long-term memory and mental imagery.
| S-EPMC8636799 | biostudies-literature
Project description:Images and videos containing fake faces are the most common type of digital manipulation. Such content can lead to negative consequences by spreading false information. The use of machine learning algorithms to produce fake face images has made it challenging to distinguish between genuine and fake content. Face manipulations are categorized into four basic groups: entire face synthesis, face identity manipulation (deepfake), facial attribute manipulation and facial expression manipulation. The study utilized lightweight convolutional neural networks to detect fake face images generated by using entire face synthesis and generative adversarial networks. The dataset used in the training process includes 70,000 real images in the FFHQ dataset and 70,000 fake images produced with StyleGAN2 using the FFHQ dataset. 80% of the dataset was used for training and 20% for testing. Initially, the MobileNet, MobileNetV2, EfficientNetB0, and NASNetMobile convolutional neural networks were trained separately for the training process. In the training, the models were pre-trained on ImageNet and reused with transfer learning. As a result of the first trainings EfficientNetB0 algorithm reached the highest accuracy of 93.64%. The EfficientNetB0 algorithm was revised to increase its accuracy rate by adding two dense layers (256 neurons) with ReLU activation, two dropout layers, one flattening layer, one dense layer (128 neurons) with ReLU activation function, and a softmax activation function used for the classification dense layer with two nodes. As a result of this process accuracy rate of 95.48% was achieved with EfficientNetB0 algorithm. Finally, the model that achieved 95.48% accuracy was used to train MobileNet and MobileNetV2 models together using the stacking ensemble learning method, resulting in the highest accuracy rate of 96.44%.
| S-EPMC11232570 | biostudies-literature