1,721,070 research outputs found
The role of artificial intelligence in implant dentistry: a systematic review
No full textThe aim of this systematic review was to comprehensively analyse recent studies on the application of artificial intelligence (AI) in dental implantology. The PRISMA guidelines were followed. Five databases were accessed: Scopus, Web of Science, MEDLINE/PubMed, IEEE Xplore, and JSTOR. Documents published between 2018 and October 15, 2024 relating to AI and implantology were considered. Exclusions encompassed reviews, opinion articles, books, conference references, studies using AI as a supplementary method, AI for teaching implant dentistry, and AI for implant fabrication, prothesis, or design. A total of 120 relevant papers were included. Risk of bias was assessed using PROBAST. Findings demonstrated extensive utilization of AI in various aspects of dental implantology: guided surgery, diagnosis, classification of oral structures, bone classification, classification of dental restorations, implant classification, implant planning, and implant prognosis. Deep learning algorithms were employed in 89.2% of studies, predominantly utilizing image data (72.0% two-dimensional images and 28.0% three-dimensional images). Publications doubled in 2022 compared to the previous year and have remained consistent since. Despite growth, the field remains relatively underdeveloped. However, with advancements in technology and data quality, substantial progress is anticipated in forthcoming years. Remarkably, 11 studies were found to have a high risk of bias.This work was supported by the BIKAINTEK program of the Basque Government under Grant no. 007-B2/2021.Peer reviewe
Folded: A toolkit to describe mammalian herbivore dentition from 2D images
No full textAbstract Dental morphology is a major aspect of ecological and evolutionary studies of both extant and fossil mammalian species. Mammalian dentitions are diverse feeding systems that can be defined through continuous numerical descriptors of the enamel pattern. We developed a comprehensive toolkit to quantify complex occlusal enamel patterns from two‐dimensional images of herbivore mammals, widespread in the scientific literature, in form of three novel enamel complexity descriptors: two‐dimensional orientation patch count (2D OPC), enamel folding (EF), and enamel thickness (ET). Previously proposed parameters such as occlusal enamel index or indentation index are implemented as well. The current method is devised for extracting continuous variables of enamel complexity from macro and microherbivore mammalian species with conspicuous wear facets. A general case study is proposed using two clades within the Family Rhinocerotidae containing species regarded as hypsodonts. The results show that antagonist dental adaptations were achieved through disparate evolutionary strategies in both groups. To test the robustness of this tool under different practical scenarios, other mammalian groups have been evaluated as well. Additional sensitivity analyses include the impact of image size, rotation, or differences in dental wear. Our approach differs from previous 2D techniques in its affordability, versatility, and control over individual regions within each tooth while delivering continuous numerical data. Additionally, the 2D reference images required as input are widespread in the literature and easier to process in comparison to 3D data alternatives
Reproyección multi-vista de objetos para vehículos autónomos
Get PDFLa detección y reproyección de objetos 3D en imágenes monoculares ha sido poco usada
en el ámbito de la conducción autónoma debido a la gran eficacia del uso de LiDARs
(Laser Imaging Detection and Ranging). El problema con este método es el precio de
estos sistemas láser. Con el aumento de la capacidad de cómputo de tanto tarjetas gráficas
como procesadores, la tarea de reproyección usando solo las detecciones de redes neuro-
nales y cámaras calibradas está volviéndose más viable. Es por esto que se ha propuesto
un método de reproyección de detecciones 2D, realizando la calibración de un sistema
de 4 cámaras en el vehículo de testeo del centro de investigación Vicomtech. El método
se ha integrado en un diagrama RTMaps (Real Time Multisensor applications) en tiempo
real, y utilizado en el vehículo de test en diferentes demostraciones. Adicionalmente, se
ha realizado un estudio del error de reproyección utilizando una base de datos anotada y
accesible públicamente
Prototipo CAD de segmentación automática de cáncer de pulmón en imágenes histopatológicas TMA
Get PDFEl cáncer de pulmón es una enfermedad letal que para el 2012 se situó como la quinta causa de muerte a nivel mundial, la tercera en Europa y la primera en España con casi 20.000 nuevos casos cada año; aproximadamente el 85 % de los sujetos que padecen cáncer de pulmón, morirán por esta enfermedad. El principal obstáculo en la lucha contra esta patología es su detección tardía. El desarrollo que ha experimentado el campo de la imagen médica en aspectos como la adquisición, almacenamiento y visualización ha contribuido al mejoramiento de la calidad del análisis y diagnóstico de las diferentes patologías (entre ellas el cáncer de pulmón) convirtiéndola actualmente en un componente indispensable en medicina. En las últimas décadas, se han realizado numerosos esfuerzos para detectar de manera precoz el cáncer de pulmón mediante el desarrollo de distintas tecnologías, entre ellas los sistemas de diagnóstico asistido por computador (CAD), los cuales mediante el análisis automático de la imagen médica brindan al especialista una segunda opinión diagnostica, con el objetivo de obtener diagnósticos mas precisos que permitan formular tratamientos mas adecuados. La imagen médica histopatológica es el "gold standard. en detección temprana de la mayoría de patológicas incluido el cáncer de pulmón. La tarea de detección suele ser bastante tediosa e que implica una importante inversión de tiempo y esfuerzo por parte de los expertos en histopatología. El crecimiento de los bancos de tejidos ya ha superado las habilidades manuales de análisis disponibles. Además, la revisión de patología experta sufre variaciones ínter e intra observador. Lo anterior evidencia la gran necesidad de automatizar el análisis de imagen médica en histopatológica. En este trabajo se hace una aproximación a la detección de cáncer de pulmón en imagen médica, concretamente abordando el problema de segmentación de tejido tumoral y no tumoral sobre imágenes histopatológicas TMA, mediante el desarrollo de un prototipo de sistema de diagnóstico asistido por computador CAD
Segmentación automática de procesos neuronales en microscopı́a electrónica mediante técnicas de aprendizaje profundo
Get PDFEn este trabajo se han utilizado redes neuronales convolucionales para la seg-
mentación de imágenes biomédicas obtenidas mediante microscopia electrónica.
El trabajo se ha desarrollado usando la librerı́a de Keras y ayudándonos
de la herramienta Google Colaboratory para la ejecución de los modelos más
pesados computacionalmente. Se ha comenzado entrenando redes neuronales
artificiales para ir adentrándonos en el funcionamiento de la librerı́a. Después
se ha entrenado una red preentrenada, concretamente la VGG16, bloqueando
todas sus capas convolucionales y dejando alguna desbloqueada. Y finalmente
se ha modelado una red neuronal convolucional siguiendo la estructura de la red
U-Net. Esta red ha dado buenos resultados en la segmentación de imágenes y
se utiliza sobre todo en la segmentación de imágenes biomédicas.
La base de datos del caso principal, se ha obtenido de la competición lanza-
da en el International Symposium on Biomedical Imaging (ISBI) de 2012 y está
compuesta por un conjunto de cortes de microscopia electrónica para entrenar
algoritmos de aprendizaje automáticos y ası́ poder realizar la segmentación au-
tomática de neuritas
Face beauty analysis via manifold based semi-supervised learning
Get PDFBeauty has always played an important role in society, implicitly influencing the hu-
man interactions of our daily lives and more significant aspects, such as the mate
choice or job interviews. And now, with the progress made in deep learning and fea-
ture extraction, automatic facial beauty analysis has become an emerging research
topic too. However, the subjectivity of beauty still hinders the developement in this
area, due to the cost of collecting reliable labeled data, since the beauty score of an
individual has to be determined according to various raters.
To address this problem, we study the performances of four different semi-supervised
manifold based algorithms, which can take advantage of both labeled and unlabeled
data in the training phase, and we use them in two different datasets: SCUT-FBP
and M 2 B. The learning algorithms are Local and Global Consistency, Flexible Man-
ifold Embedding and Kernel Flexible Manifold Embedding. There is an additional
algorithm, which, unlike the rest of them, instead of performing classification, ob-
tains a non-linear transformation of the data to make the classification easier. All of
these algorithms were designed to work on discrete classes, but we perform regres-
sion, where labels are real numbers. So the first step, in chapter 2, is to analyse how
the algorithms can be adapted to regression and to hypothesize which problems we
could be encountering in this process. Secondly, we empirically test them (chapter
3). The best results are obtained with KFME on both datasets, achieving a mean
average error of 0.0104 (out of 1) and a Pearson correlation of 0.9782 on SCUT-FBP
dataset. With respect to M 2 B dataset, a mean average error of 0.0697 and a Pear-
son correlation of 0.7757 are achieved on eastern faces, while a mean average error
of 0.0717 and a Pearson correlation of 0.7848 are achieved on western faces. This
dissertation ends with a final chapter discussing the results and proposing new topics
of study for future work
Image-based family verification in the wild
Get PDFFacial image analysis has been an important subject of study in the communities of pat-
tern recognition and computer vision. Facial images contain much information about the
person they belong to: identity, age, gender, ethnicity, expression and many more. For that
reason, the analysis of facial images has many applications in real world problems such
as face recognition, age estimation, gender classification or facial expression recognition.
Visual kinship recognition is a new research topic in the scope of facial image analysis.
It is essential for many real-world applications. However, nowadays
there exist only a few practical vision systems capable to handle such tasks. Hence, vision
technology for kinship-based problems has not matured enough to be applied to real-
world problems. This leads to a concern of unsatisfactory performance when attempted
on real-world datasets.
Kinship verification is to determine pairwise kin relations for a pair of given images. It
can be viewed as a typical binary classification problem, i.e., a face pair is either related
by kinship or it is not. Prior research works have addressed kinship types
for which pre-existing datasets have provided images, annotations and a verification task
protocol. Namely, father-son, father-daughter, mother-son and mother-daughter.
The main objective of this Master work is the study and development of feature selection
and fusion for the problem of family verification from facial images.
To achieve this objective, there is a main tasks that can be addressed: perform a compara-
tive study on face descriptors that include classic descriptors as well as deep descriptors.
The main contributions of this Thesis work are:
1. Studying the state of the art of the problem of family verification in images.
2. Implementing and comparing several criteria that correspond to different face rep-
resentations (Local Binary Patterns (LBP), Histogram Oriented Gradients (HOG),
deep descriptors)
Driver drowsiness detection in facial images
Get PDFDriver fatigue is a significant factor in a large number of vehicle accidents. Thus, drowsy
driver alert systems are meant to reduce the main cause of traffic accidents. Different
approaches have been developed to tackle with the fatigue detection problem. Though
most reliable techniques to asses fatigue involve the use of physical sensors to monitor
drivers, they can be too intrusive and are less likely to be adopted by the car industry. A
relatively new and effective trend consists on facial image analysis from video cameras
that monitor drivers.
How to extract effective features of fatigue from images is important for many image
processing applications. This project proposes a face descriptor that can be used to detect
driver fatigue in static frames. This descriptor represents each frame of a sequence as
a pyramid of scaled images that are divided into non-overlapping blocks of equal size.
The pyramid of images is combined with three different image descriptors. The final
descriptors are filtered out using feature selection and a Support Vector Machine is used
to predict the drowsiness state. The proposed method is tested on the public NTHUDDD
dataset, which is the state-of-the-art dataset on driver drowsiness detection
Analysis of facial expressions: experiments on multiple databases
Get PDFThis master thesis compares different face descriptors using classification techniques in
order to classify emotions in images of faces of people of different ethnicities and ages,
male and female. The comparison is done between hand-crafted features such as LBP and
HOG and more modern features such as some pre-trained neural networks. The proposed
methods were used on different databases, using different image sizes and cropping and
standardizing all the images. The experimental results showed that some of the hand-
crafted features were better that the pre-trained neural networks. To facilitate replication
of our experiments the MATLBAB source code will be available at https://github.
com/nagwlei/FaceEmotions
Advancing artificial intelligence in medical imaging: self- aware exploration, data augmentation, and real- world simulation for robust diagnosis
No full text213 p.Artificial intelligence (Al) and deep learning are at the forefront of revolutionizing medica! imaging, promising unprecedented advancements in diagnostic precision, efficiency, and accessibility. Despite this potential, their application in clinical settings is hampered by persistent challenges such as limited high-quality data, class imbalance, and the complex variability inherent in medica! images. This PhD thesis addresses these critica! challenges through three pioneering research contributions: (1) the introduction of self-aware Al systems capable of autonomous learning optimization, (2) the creation of a novel data augmentation technique that enhances model performance under constrained data conditions, and (3) the development of robust deep learning frameworks tailored for MRl based brain cancer diagnosis in real-world scenarios.The first key innovation is the exploration of self-aware Al. This thesis propases that pre-trained models such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) possess the potential to self-monitor and adapt their learning processes autonomously. By introducing novel metrics, such as "accuracy difference" and "loss difference," this research offers a framework for measuring and enhancing the self-awareness of Al models. This development has profound implications for the future of Al in healthcare, paving the way for adaptive models that can fine-tune themselves in response to the quality and quantity of the datathey encounter, ensuring more reliable and context-aware decision-making in clinical environments.The second contribution is the introduction of the 'Naturalize' augmentation technique, a groundbreaking method designed to generate synthetic medical images with fidelity equal to original data. This technique is specifically engineered to overcome data scarcity and class imbalance, two significant obstacles in medical imaging. By applying 'Naturalize' to blood cell and skin cancer classification, this work demonstrates unparalleled improvements in both sensitivity and specificity, particularly in underrepresented classes. The 'Naturalize' technique fundamentally transforms the dataset augmentation paradigm, enabling deep models to generalize more effectively to unseen clinical cases.The third contribution focuses on advancing model robustness for MRl-based brain cancer diagnosis. Medical imaging, particularly MRI, is subject to significant variability dueto differences in magnetic field strength, patient motion, and image quality. This work systematically addresses these challenges by simulating real world conditions-such as noise, blur, and motion artifacts-within the training process. By integrating sophisticated data augmentation strategies, including Gaussian noise and blur, into the training of deep learning models, the research significantly enhances model resilience and generalization, ensuring consistent and accurate performance across diverse clinical environments. This contribution is a critica! step toward translating Al solutions from research to practice in the domain of neuro-oncology.In summary, this thesis makes substantial strides in addressing sorne of the most pressing limitations of Al in medical imaging. Through innovative augmentation techniques, the concept of self-aware Al, and robust training methodologies, this work significantly enhances the reliability, adaptability, and clinical relevance of deep learning models. The findings lay the foundation for more intelligent, autonomous, and clinically applicable Al systems that can improve diagnostic accuracy and patient outcomes. Future research directions include expanding the 'Naturalize' method to 3D imaging, developing real-time adaptive Al models, and exploring federated learning to enhance model generalization while preserving patient privacy.K.eywords: Deep Learning, Transfer Learning, Fine-tuning, Medica! lmage Augmentation, Naturalize, GANs, Pre-trained Models, Medical lmaging, Self awareness, Model Robustness, Real-Life Scenario Simulatio
- …
