Computer-aided autism diagnosis using visual attention models and eye-tracking: replication and improvement proposal.

BACKGROUND: Autism Spectrum Disorder (ASD) diagnosis can be aided by approaches based on eye-tracking signals. Recently, the feasibility of building Visual Attention Models (VAMs) from features extracted from visual stimuli and their use for classifying cases and controls has been demonstrated using Neural Networks and Support Vector Machines. The present work has three aims: 1) to evaluate whether the trained classifier from the previous study was generalist enough to classify new samples with a new stimulus; 2) to replicate the previously approach to train a new classifier with a new dataset; 3) to evaluate the performance of classifiers obtained by a new classification algorithm (Random Forest) using the previous and the current datasets. METHODS: The previously approach was replicated with a new stimulus and new sample, 44 from the Typical Development group and 33 from the ASD group. After the replication, Random Forest classifier was tested to substitute Neural Networks algorithm. RESULTS: The test with the trained classifier reached an AUC of 0.56, suggesting that the trained classifier requires retraining of the VAMs when changing the stimulus. The replication results reached an AUC of 0.71, indicating the potential of generalization of the approach for aiding ASD diagnosis, as long as the stimulus is similar to the originally proposed. The results achieved with Random Forest were superior to those achieved with the original approach, with an average AUC of 0.95 for the previous dataset and 0.74 for the new dataset. CONCLUSION: In summary, the results of the replication experiment were satisfactory, which suggests the robustness of the approach and the VAM-based approaches feasibility to aid in ASD diagnosis. The proposed method change improved the classification performance. Some limitations are discussed and additional studies are encouraged to test other conditions and scenarios.

Computer aided diagnosis of neurodevelopmental disorders and genetic syndromes based on facial images - A systematic literature review.

Neurodevelopment disorders can result in facial dysmorphisms. Therefore, the analysis of facial images using image processing and machine learning techniques can help construct systems for diagnosing genetic syndromes and neurodevelopmental disorders. The systems offer faster and cost-effective alternatives for genotyping tests, particularly when dealing with large-scale applications. However, there are still challenges to overcome to ensure the accuracy and reliability of computer-aided diagnosis systems. This article presents a systematic review of such initiatives, including 55 articles. The main aspects used to develop these diagnostic systems were discussed, namely datasets - availability, type of image, size, ethnicities and syndromes - types of facial features, techniques used for normalization, dimensionality reduction and classification, deep learning, as well as a discussion related to the main gaps, challenges and opportunities.

A Stochastic Grammar Approach to Mass Classification in Mammograms.

Breast cancer is responsible for approximately 15% of all cancer-related deaths among women worldwide, and early and accurate diagnosis increases the chances of survival. Over the last decades, several machine learning approaches have been used to improve the diagnosis of this disease, but most of them require a large set of samples for training. Syntactic approaches were barely used in this context, although it can present good results even if the training set has few samples. This article presents a syntactic approach to classify masses as benign or malignant. There were used features extracted from a polygonal representation of masses combined with a stochastic grammar approach to discriminate the masses found in mammograms. The results were compared with other machine learning techniques, and the grammar-based classifiers showed superior performance in the classification task. The best accuracies achieved were from 96% to 100%, indicating that grammatical approaches are robust and able to discriminate the masses even when trained with small samples of images. Syntactic approaches could be more frequently employed in the classification of masses, since they can learn the pattern of benign and malignant masses from a small sample of images achieving similar results when compared to the state of art.

Correlating drug prescriptions with prognosis in severe COVID-19: first step towards resource management.

BACKGROUND: Optimal COVID-19 management is still undefined. In this complicated scenario, the construction of a computational model capable of extracting information from electronic medical records, correlating signs, symptoms and medical prescriptions, could improve patient management/prognosis. METHODS: The aim of this study is to investigate the correlation between drug prescriptions and outcome in patients with COVID-19. We extracted data from 3674 medical records of hospitalized patients: drug prescriptions, outcome, and demographics. The outcome evaluated was hospital outcome. We applied correlation analysis using a Logistic Regression algorithm for machine learning with Lasso and Matthews correlation coefficient. RESULTS: We found correlations between drugs and patient outcomes (death/discharged alive). Anticoagulants, used very frequently during all phases of the disease, were associated with good prognosis only after the first week of symptoms. Antibiotics very frequently prescribed, especially early, were not correlated with outcome, suggesting that bacterial infections may not be important in determining prognosis. There were no differences between age groups. CONCLUSIONS: In conclusion, we achieved an important result in the area of Artificial Intelligence, as we were able to establish a correlation between concrete variables in a real and extremely complex environment of clinical data from COVID-19. Our results are an initial and promising contribution in decision-making and real-time environments to support resource management and forecasting prognosis of patients with COVID-19.

Predicting the outcome for COVID-19 patients by applying time series classification to electronic health records.

BACKGROUND: COVID-19 caused more than 622 thousand deaths in Brazil. The infection can be asymptomatic and cause mild symptoms, but it also can evolve into a severe disease and lead to death. It is difficult to predict which patients will develop severe disease. There are, in the literature, machine learning models capable of assisting diagnose and predicting outcomes for several diseases, but usually these models require laboratory tests and/or imaging. METHODS: We conducted a observational cohort study that evaluated vital signs and measurements from patients who were admitted to Hospital das Clínicas (São Paulo, Brazil) between March 2020 and October 2021 due to COVID-19. The data was then represented as univariate and multivariate time series, that were used to train and test machine learning models capable of predicting a patient's outcome. RESULTS: Time series-based machine learning models are capable of predicting a COVID-19 patient's outcome with up to 96% general accuracy and 81% accuracy considering only the first hospitalization day. The models can reach up to 99% sensitivity (discharge prediction) and up to 91% specificity (death prediction). CONCLUSIONS: Results indicate that time series-based machine learning models combined with easily obtainable data can predict COVID-19 outcomes and support clinical decisions. With further research, these models can potentially help doctors diagnose other diseases.

A systematic review of multi-slice and multi-frame descriptors in cardiac MRI exams.

Computer-Aided Diagnosis systems have been developed to help medical professional in their decision making routines towards a more accurate diagnosis. These systems process medical exams such as Magnetic Resonance (MRI) in order to quantify meaningful features. These can be used with similarity-measuring techniques in a Content-Based Image Retrieval context, or inputted into a machine learning classifier in order to support early disease detection. For cardiac MRIs, single slice descriptors have been proposed in the two-dimensional domain, shape descriptors have been proposed in the three-dimensional domain, and previous reviews have mapped these two descriptor categories. Nonetheless, no systematic review on these descriptors have looked at full cardiac MRI images sets. We have reviewed the literature by searching for descriptors that consider the whole slice set (multi-slice) or frames (multi-frame) in cardiac MRI exams. We discuss descriptors and techniques, the datasets that were used, and the different evaluation metrics. Finally, we highlight literature gaps and research opportunities.

Can Individuals Poststroke Improve Their Performance in Reaction and Movement Times in a Nonimmersive Serious Game with Practice? A Cross-Sectional Study.

Objective: To verify if individuals' poststroke and healthy controls would improve their performance in reaction and movement times practicing a serious game task using the upper limb movements. Materials and Methods: We evaluated 30 individuals poststroke and 30 healthy controls, matched for age and sex. We used the "Association Game for Rehabilitation" (AGaR) where participants played by matching a pair of images whose meanings were similar. Hand movements were captured by a Kinect system and poststroke participants used their nonparetic upper limb. Reaction time and movement times (time to select an image and movement time to the target) were measured. Data were analyzed using multiple analysis of variance. Results: Performance improved for both groups across all variables with better performance in movement times than reaction time only for poststroke individuals. Conclusions: Upper limb movements using nonimmersive serious games enhanced motor performance in reaction and movement times for healthy controls and individuals poststroke. ReBEC Trial Registration: RBR-4m4pk; Registeted on 08/24/2018.

Computer-aided autism diagnosis based on visual attention models using eye tracking.

An advantage of using eye tracking for diagnosis is that it is non-invasive and can be performed in individuals with different functional levels and ages. Computer/aided diagnosis using eye tracking data is commonly based on eye fixation points in some regions of interest (ROI) in an image. However, besides the need for every ROI demarcation in each image or video frame used in the experiment, the diversity of visual features contained in each ROI may compromise the characterization of visual attention in each group (case or control) and consequent diagnosis accuracy. Although some approaches use eye tracking signals for aiding diagnosis, it is still a challenge to identify frames of interest when videos are used as stimuli and to select relevant characteristics extracted from the videos. This is mainly observed in applications for autism spectrum disorder (ASD) diagnosis. To address these issues, the present paper proposes: (1) a computational method, integrating concepts of Visual Attention Model, Image Processing and Artificial Intelligence techniques for learning a model for each group (case and control) using eye tracking data, and (2) a supervised classifier that, using the learned models, performs the diagnosis. Although this approach is not disorder-specific, it was tested in the context of ASD diagnosis, obtaining an average of precision, recall and specificity of 90%, 69% and 93%, respectively.

Visual Search Efficiency in Mild Cognitive Impairment and Alzheimer's Disease: An Eye Movement Study.

BACKGROUND: Visual search abilities are essential to everyday life activities and are known to be affected in Alzheimer's disease (AD). However, little is known about visual search efficiency in mild cognitive impairment (MCI), a transitive state between normal aging and dementia. Eye movement studies and machine learning methods have been recently used to detect oculomotor impairments in individuals with dementia. OBJECTIVE: The aim of the present study is to investigate the association between eye movement metrics and visual search impairment in MCI and AD. METHODS: 127 participants were tested: 43 healthy controls, 51 with MCI, and 33 with AD. They completed an eyetracking visual search task where they had to find a previously seen target stimulus among distractors. RESULTS: Both patient groups made more fixations on the screen when searching for a target, with longer duration than controls. MCI and AD fixated the distractors more often and for a longer period of time than the target. Healthy controls were quicker and made less fixations when scanning the stimuli for the first time. Machine-learning methods were able to distinguish between controls and AD subjects and to identify MCI subjects with a similar oculomotor profile to AD with a good accuracy. CONCLUSION: Results showed that eye movement metrics are useful for identifying visual search impairments in MCI and AD, with possible implications in the early identification of individuals with high-risk of developing AD.

Haptic interaction for needle insertion training in medical applications: The state-of-the-art.

Computer-based simulation for medical procedures training has been gaining relevance, as well the use of haptic devices for developing fine motor skills in such simulations. The purpose of this paper is to present a review of the state-of-the-art in virtual needle insertion training simulation based on haptic interaction. A systematic review method was applied to gather documentation that enables a rigorous audit of the process stages and results. We established a classification system based on certain characteristics of the studies analyzed, including: main procedures and target body regions in medical applications; ways to generate haptic feedback; devices; types of environment; and user validation. In addition, the review aimed to identify challenges and trends in the field, indicating research opportunities. Results showed the predominance of Virtual Reality and commercial haptic devices in simulations. Since most studies are based on subjective tests, finding ways to objectively evaluate haptic interaction perception represents a promising research field. We also found that devices and ways to generate haptic feedback and to represent tissue and needle behavior pose limitations and challenges for computer simulation. Finally, the realism provided is a constant concern in the validation process, which brings another problem: defining and performing suitable user tests.

Achievement of Virtual and Real Objects Using a Short-Term Motor Learning Protocol in People with Duchenne Muscular Dystrophy: A Crossover Randomized Controlled Trial.

OBJECTIVE: To evaluate whether people with Duchenne muscular dystrophy (DMD) practicing a task in a virtual environment could improve performance given a similar task in a real environment, as well as distinguishing whether there is transference between performing the practice in virtual environment and then a real environment and vice versa. METHODS: Twenty-two people with DMD were evaluated and divided into two groups. The goal was to reach out and touch a red cube. Group A began with the real task and had to touch a real object, and Group B began with the virtual task and had to reach a virtual object using the Kinect system. RESULTS: ANOVA showed that all participants decreased the movement time from the first (M = 973 ms) to the last block of acquisition (M = 783 ms) in both virtual and real tasks and motor learning could be inferred by the short-term retention and transfer task (with increasing distance of the target). However, the evaluation of task performance demonstrated that the virtual task provided an inferior performance when compared to the real task in all phases of the study, and there was no effect for sequence. CONCLUSIONS: Both virtual and real tasks promoted improvement of performance in the acquisition phase, short-term retention, and transfer. However, there was no transference of learning between environments. In conclusion, it is recommended that the use of virtual environments for individuals with DMD needs to be considered carefully.

Comparing efficient data structures to represent geometric models for three-dimensional virtual medical training.

Data structures have been explored for several domains of computer applications in order to ensure efficiency in the data store and retrieval. However, data structures can present different behavior depending on applications that they are being used. Three-dimensional interactive environments offered by techniques of Virtual Reality require operations of loading and manipulating objects in real time, where realism and response time are two important requirements. Efficient representation of geometrical models plays an important part so that the simulation may become real. In this paper, we present the implementation and the comparison of two topologically efficient data structures - Compact Half-Edge and Mate-Face - for the representation of objects for three-dimensional interactive environments. The structures have been tested at different conditions of processors and RAM memories. The results show that both these structures can be used in an efficient manner. Mate-Face structure has shown itself to be more efficient for the manipulation of neighborhood relationships and the Compact Half-Edge was more efficient for loading of the geometric models. We also evaluated the data structures embedded in applications of biopsy simulation using virtual reality, considering a deformation simulation method applied in virtual human organs. The results showed that their use allows the building of applications considering objects with high resolutions (number of vertices), without significant impact in the time spent in the simulation. Therefore, their use contributes for the construction of more realistic simulators.

Three-dimensional Content-Based Cardiac Image Retrieval using global and local descriptors.

The increase in volume of medical images generated and stored has created difficulties in accurate image retrieval. An alternative is to generate three-dimensional (3D) models from such medical images and use them in the search. Some of the main cardiac illnesses, such as Congestive Heart Failure (CHF), have deformation in the heart's shape as one of the main symptoms, which can be identified faster in a 3D object than in slices. This article presents techniques developed to retrieve 3D cardiac models using global and local descriptors within a content-based image retrieval system. These techniques were applied in pre-classified 3D models with and without the CHF disease and they were evaluated by using Precision vs. Recall metric. We observed that local descriptors achieved better results than a global descriptor, reaching 85% of accuracy. The results confirmed the potential of using 3D models retrieval in the medical context to aid in the diagnosis.

Applying Distance Histogram to retrieve 3D cardiac medical models.

Three-dimensional models are being extensively used in the medical area in order to improve clinical medical examinations and diagnosis. The Cardiology field handles with several types of image slices to compose the diagnosis. MRI (Magnetic Resonance Imaging) is a non-invasive technique to detect anomalies from internal images of the human body that generates hundreds of images, which takes long for the specialist to analyze frame by frame and the diagnosis precision can be affected. Many cardiac diseases could be identified through shape deformation, but systems aimed to aid diagnosis usually identify shapes in two-dimensional (2D) images. Our aim is to apply a shape descriptor to retrieve three-dimensional cardiac models, obtained from a set of 2D slices, which were segmented and reconstructed from MRI images using their geometry information. Preliminary results show that the shape deformation in 3D models can be a good indicator to detect Congestive Heart Failure, a very common heart disease.

Contrast enhancement in dense breast images to aid clustered microcalcifications detection.

This paper presents a method to provide contrast enhancement in dense breast digitized images, which are difficult cases in testing of computer-aided diagnosis (CAD) schemes. Three techniques were developed, and data from each method were combined to provide a better result in relation to detection of clustered microcalcifications. Results obtained during the tests indicated that, by combining all the developed techniques, it is possible to improve the performance of a processing scheme designed to detect microcalcification clusters. It also allows operators to distinguish some of these structures in low-contrast images, which were not detected via conventional processing before the contrast enhancement. This investigation shows the possibility of improving CAD schemes for better detection of microcalcifications in dense breast images.

Evaluation of microcalcifications segmentation techniques for dense breast digitized images.

Some processing techniques cited in the literature used for microcalcifications detection in digitized mammograms are evaluated here with regard to dense breast images. Three techniques were investigated: Nappi et al.'s, Nishikawa et al.'s and Wallet et al.'s. The methods were tested with low-contrast phantom images, simulating dense breast images. The ability of each technique to detect microcalcifications in dense breast images was evaluated. The following detection rates were obtained: Nappi et al's technique, 78.3%; Wallet et al.'s, 86.6%; and Nishikawa et al.'s, 94.4%. Dense breast images affect the performance of CAD schemes, as confirmed by our results. Therefore, data from those segmentation techniques applied to dense breast images could be improved by developing a hybrid method using the best characteristics of each technique.

Contrast enhancement in dense breast images using the modulation transfer function.

This work proposes a method aimed at enhancing the contrast in dense breast images in mammography. It includes a new preprocessing technique, which uses information on the modulation transfer function (MTF) of the mammographic system in the whole radiation field. The method is applied to improve the efficiency of a computer-aided diagnosis (CAD) scheme. Seventy-five regions of interest (ROIs) from dense mammograms were acquired in two pieces of equipment (a CGR Senographe 500t and a Philips Mammodiagnost) and were digitized in a Lumiscan 50 laser scanner. A computational procedure determines the effective focal spot size in each region of interest from the measured focal spot in the center for a given mammographic equipment. Using computational simulation the MTF is then calculated for each field region. A procedure that enlarges the high-frequency portion of this function is applied and a convolution between the resulting new function and the original image is performed. Both original and enhanced images were submitted to a processing procedure for detecting clustered microcalcifications in order to compare the performance for dense breast images. ROIs were divided into four groups, two for each piece of equipment-one with clustered microcalcifications and another without microcalcifications. Our results show that in about 10% of the enhanced images more signals were detected when compared to the results for the original dense breast images. This is important because the usual processing techniques used in CAD schemes present poor results when applied to dense breast images. Since the MTF method is a well-recognized tool in the evaluation of radiographic systems, this new technique could be used to associate quality assurance procedures with the processing schemes employed in CAD for mammography.

Detecção de agrupamentos de microcalcificações mamárias através de processamento de imagens mamográficas com detecção semi-automática de regiões de interesse / Breast clustered microcalcifications detection by mammographic image processing with regions of interest semi-automatic detection

A necessidade de detecção precoce do câncer de mama tem levado centros de pesquisa ao desenvolvimento de esquemas de diagnóstico auxiliado por computador, pois, se diagnosticado na fase inicial do desenvolvimento, esse tipo de câncer apresenta grandes chances de cura. Este trabalho apresenta um esquema computacional para diagnosticar um dos indícios da possível existência de um tumor: os agrupamento ("clusters") de microcalcificações mamárias. Foi desenvolvido um método de detecção semi-automática das regiões de interesse em mamogramas digitalizados. O sistema utiliza a técnica de transformação área-ponto e possibilita a identificação dos clusters através de dois métodos de processamento: mascaramento e crescimento de região,além de indicar a localização dos clusters detectados. Os testes indicaram um acerto de 94 por cento na identificação das aglomerações em um particular conjunto de mamogramas reais.