Evaluating artificial intelligence for comparative radiography.

INTRODUCTION: Comparative radiography is a forensic identification and shortlisting technique based on the comparison of skeletal structures in ante-mortem and post-mortem images. The images (e.g., 2D radiographs or 3D computed tomographies) are manually superimposed and visually compared by a forensic practitioner. It requires a significant amount of time per comparison, limiting its utility in large comparison scenarios. METHODS: We propose and validate a novel framework for automating the shortlisting of candidates using artificial intelligence. It is composed of (1) a segmentation method to delimit skeletal structures' silhouettes in radiographs, (2) a superposition method to generate the best simulated "radiographs" from 3D images according to the segmented radiographs, and (3) a decision-making method for shortlisting all candidates ranked according to a similarity metric. MATERIAL: The dataset is composed of 180 computed tomographies and 180 radiographs where the frontal sinuses are visible. Frontal sinuses are the skeletal structure analyzed due to their high individualization capability. RESULTS: Firstly, we validate two deep learning-based techniques for segmenting the frontal sinuses in radiographs, obtaining high-quality results. Secondly, we study the framework's shortlisting capability using both automatic segmentations and superimpositions. The obtained superimpositions, based only on the superimposition metric, allowed us to filter out 40% of the possible candidates in a completely automatic manner. Thirdly, we perform a reliability study by comparing 180 radiographs against 180 computed tomographies using manual segmentations. The results allowed us to filter out 73% of the possible candidates. Furthermore, the results are robust to inter- and intra-expert-related errors.

Analysis of the performance of machine learning and deep learning methods for sex estimation of infant individuals from the analysis of 2D images of the ilium.

Reducing the subjectivity of the methods used for biological profile estimation is, at present, a priority research line in forensic anthropology. To achieve this, artificial intelligence (AI) techniques can be a valuable tool yet to be exploited in this discipline. The goal of this study is to compare the effectiveness of different machine learning (ML) methods with the visual assessment of an expert to estimate the sex of infant skeletons from images of the ilium. Photographs of the ilium of 135 individuals, age between 5 months of gestation and 6 years, from the collection of identified infant skeletons of the University of Granada have been used, and classic ML and deep learning (DL) techniques have been applied to develop prediction algorithms. To assess their effectiveness, the results have been compared with those obtained by a forensic expert, who has estimated the sex from each photograph through direct observation and subjective assessment following the criteria described by Schutkowsky in 1993. The results show that the algorithms obtained using DL techniques offer an accuracy of 59%, very close to the 61% obtained by the expert, and 10 percentual points better than classic ML techniques. This study offers promising results and represents the first AI-based approach for estimating sex in infant individuals using photographs of the ilium.

Age estimation in forensic anthropology: methodological considerations about the validation studies of prediction models.

There is currently no clear consensus on how to calculate, express, and interpret the error when validating methods for age estimation in forensic anthropology. For this reason, it is likely that researchers are commonly drawing erroneous or confusing conclusions about the existence of population differences or the need to design new and increasingly complex estimation methods. In recent years, many researchers have highlighted these limitations. They propose new lines of research focused on the use of rigorous statistics and new technologies for the development of methods for estimating age. Our main objective in this study is to contribute to the strengthening of these novel ideas, for which we show the existing empirical evidence about the inadequacy of some age estimation methods in calculating, expressing, and interpreting the errors obtained. With this aim, a total of 500 simulations have been performed, in which hypothetical research teams develop and validate methods for age estimation. The data employed in this study was obtained from the "Centers for Disease Control and Prevention (CDC) Growth Charts: United States" released in 2000. The charts relate age with height, weight, and head circumference of US male children. Five learning algorithms have been employed as age estimators. We have performed three experiments in which the following aspects have been analyzed: frequency with which "negative" results can be obtained in the validation studies; which are the most appropriate criteria to compare and select the age estimation methods; and what analysis should be employed to carry out the validation studies. The results show possible errors in the interpretation of validation studies as a consequence of the confusion of statistical concepts. To conclude, we made a proposal of "good practices" for the correct calculation, expression, and interpretation of the error when validating age estimation methods in forensic anthropology.

Information fusion for infant age estimation from deciduous teeth using machine learning.

OBJECTIVES: Over the past few years, several methods have been proposed to improve the accuracy of age estimation in infants with a focus on dental development as a reliable marker. However, traditional approaches have limitations in efficiently combining information from different teeth and features. In order to address these challenges, this article presents a study on age estimation in infants with Machine Learning (ML) techniques, using deciduous teeth. MATERIALS AND METHODS: The involved dataset comprises 114 infant skeletons from the Granada osteological collection of identified infants, aged between 5 months of gestation and 3 years of age. The samples consist of features such as the maximum length and mineralization and alveolar stages of teeth. For the purpose of designing a method capable of combining all the information available from each individual, a Multilayer Perceptron model is proposed, one of the most popular artificial neural networks. This model has been validated using the leave-one-out experimental validation protocol. Through different groups of experiments, the study examines the informativeness of the aforementioned features, individually and in combination. RESULTS: The results indicate that the fusion of different variables allows for more accurate age estimates (RMSE = 66 days) than when variables are analyzed separately (RMSE = 101 days). Additionally, the study demonstrates the benefits of involving multiple teeth, which significantly reduces the RMSE compared to a single tooth. DISCUSSION: This article underlines the clear advantages of ML-based methods, emphasizing their potential to improve the accuracy and robustness when estimating the age of infants.

Automatic landmark annotation in 3D surface scans of skulls: Methodological proposal and reliability study.

BACKGROUND AND OBJECTIVES: Craniometric landmarks are essential in many biomedical applications, such as morphometric analysis or forensic identification. The process of locating landmarks is usually a manual and slow task, highly influenced by fatigue, skills and the experience of the practitioner. Localization errors are propagated and magnified in subsequent steps, which can result in incorrect measurements or assumptions. Thereby, standardization, reliability and reproducibility lay the foundations for the necessary accuracy in subsequent measurements or anatomical analysis. In this paper, we present an automatic method to annotate 3D surface skull models taking into account anatomical and geometrical features. METHODS: The proposed method follows a hybrid structure where a deformable template is used to initialize the landmark positions. Then, a refinement stage is applied using prior anatomical knowledge to ensure a correct placement. Our proposal is validated over thirty 3D skull scans of male Caucasians, acquired by hand-held surface scanning, and a set of 58 craniometric landmarks. A statistical analysis was carried out to analyze the inter- and intra-observer variability of manual annotations and the automatic results, along with a visual assessment of the final results. RESULTS: Inter-observer errors show significant differences, which are reflected in the expert consensus used as reference. The average localization error was 2.19±1.5 mm when comparing the automatic landmarks to the reference location. The subsequent visual analysis confirmed the reliability of the refinement method for most landmarks. CONCLUSIONS: Repeated manual annotations show a high variability depending on both skills and expertise of the observer, and landmarks' location and characteristics. In contrast, the automatic method provides an accurate, robust and reproducible alternative to the tedious and error-prone task of manual landmarking.

A Comprehensive Analysis of Deep Regression.

Deep learning revolutionized data science, and recently its popularity has grown exponentially, as did the amount of papers employing deep networks. Vision tasks, such as human pose estimation, did not escape from this trend. There is a large number of deep models, where small changes in the network architecture, or in the data pre-processing, together with the stochastic nature of the optimization procedures, produce notably different results, making extremely difficult to sift methods that significantly outperform others. This situation motivates the current study, in which we perform a systematic evaluation and statistical analysis of vanilla deep regression, i.e., convolutional neural networks with a linear regression top layer. This is the first comprehensive analysis of deep regression techniques. We perform experiments on four vision problems, and report confidence intervals for the median performance as well as the statistical significance of the results, if any. Surprisingly, the variability due to different data pre-processing procedures generally eclipses the variability due to modifications in the network architecture. Our results reinforce the hypothesis according to which, in general, a general-purpose network (e.g., VGG-16 or ResNet-50) adequately tuned can yield results close to the state-of-the-art without having to resort to more complex and ad-hoc regression models.

Computer-Aided Classification of Gastrointestinal Lesions in Regular Colonoscopy.

We have developed a technique to study how good computers can be at diagnosing gastrointestinal lesions from regular (white light and narrow banded) colonoscopic videos compared to two levels of clinical knowledge (expert and beginner). Our technique includes a novel tissue classification approach which may save clinician's time by avoiding chromoendoscopy, a time-consuming staining procedure using indigo carmine. Our technique also discriminates the severity of individual lesions in patients with many polyps, so that the gastroenterologist can directly focus on those requiring polypectomy. Technically, we have designed and developed a framework combining machine learning and computer vision algorithms, which performs a virtual biopsy of hyperplastic lesions, serrated adenomas and adenomas. Serrated adenomas are very difficult to classify due to their mixed/hybrid nature and recent studies indicate that they can lead to colorectal cancer through the alternate serrated pathway. Our approach is the first step to avoid systematic biopsy for suspected hyperplastic tissues. We also propose a database of colonoscopic videos showing gastrointestinal lesions with ground truth collected from both expert image inspection and histology. We not only compare our system with the expert predictions, but we also study if the use of 3D shape features improves classification accuracy, and compare our technique's performance with three competitor methods.

Biomedical image segmentation using geometric deformable models and metaheuristics.

This paper describes a hybrid level set approach for medical image segmentation. This new geometric deformable model combines region- and edge-based information with the prior shape knowledge introduced using deformable registration. Our proposal consists of two phases: training and test. The former implies the learning of the level set parameters by means of a Genetic Algorithm, while the latter is the proper segmentation, where another metaheuristic, in this case Scatter Search, derives the shape prior. In an experimental comparison, this approach has shown a better performance than a number of state-of-the-art methods when segmenting anatomical structures from different biomedical image modalities.

Artificial neuron-glia networks learning approach based on cooperative coevolution.

Artificial Neuron-Glia Networks (ANGNs) are a novel bio-inspired machine learning approach. They extend classical Artificial Neural Networks (ANNs) by incorporating recent findings and suppositions about the way information is processed by neural and astrocytic networks in the most evolved living organisms. Although ANGNs are not a consolidated method, their performance against the traditional approach, i.e. without artificial astrocytes, was already demonstrated on classification problems. However, the corresponding learning algorithms developed so far strongly depends on a set of glial parameters which are manually tuned for each specific problem. As a consequence, previous experimental tests have to be done in order to determine an adequate set of values, making such manual parameter configuration time-consuming, error-prone, biased and problem dependent. Thus, in this paper, we propose a novel learning approach for ANGNs that fully automates the learning process, and gives the possibility of testing any kind of reasonable parameter configuration for each specific problem. This new learning algorithm, based on coevolutionary genetic algorithms, is able to properly learn all the ANGNs parameters. Its performance is tested on five classification problems achieving significantly better results than ANGN and competitive results with ANN approaches.

Visual search of neuropil-enriched RNAs from brain in situ hybridization data through the image analysis pipeline hippo-ATESC.

MOTIVATION: RNA molecules specifically enriched in the neuropil of neuronal cells and in particular in dendritic spines are of great interest for neurobiology in virtue of their involvement in synaptic structure and plasticity. The systematic recognition of such molecules is therefore a very important task. High resolution images of RNA in situ hybridization experiments contained in the Allen Brain Atlas (ABA) represent a very rich resource to identify them and have been so far exploited for this task through human-expert analysis. However, software tools that may automatically address the same objective are not very well developed. RESULTS: In this study we describe an automatic method for exploring in situ hybridization data and discover neuropil-enriched RNAs in the mouse hippocampus. We called it Hippo-ATESC (Automatic Texture Extraction from the Hippocampal region using Soft Computing). Bioinformatic validation showed that the Hippo-ATESC is very efficient in the recognition of RNAs which are manually identified by expert curators as neuropil-enriched on the same image series. Moreover, we show that our method can also highlight genes revealed by microdissection-based methods but missed by human visual inspection. We experimentally validated our approach by identifying a non-coding transcript enriched in mouse synaptosomes. The code is freely available on the web at http://ibislab.ce.unipr.it/software/hippo/.

Artificial astrocytes improve neural network performance.

Compelling evidence indicates the existence of bidirectional communication between astrocytes and neurons. Astrocytes, a type of glial cells classically considered to be passive supportive cells, have been recently demonstrated to be actively involved in the processing and regulation of synaptic information, suggesting that brain function arises from the activity of neuron-glia networks. However, the actual impact of astrocytes in neural network function is largely unknown and its application in artificial intelligence remains untested. We have investigated the consequences of including artificial astrocytes, which present the biologically defined properties involved in astrocyte-neuron communication, on artificial neural network performance. Using connectionist systems and evolutionary algorithms, we have compared the performance of artificial neural networks (NN) and artificial neuron-glia networks (NGN) to solve classification problems. We show that the degree of success of NGN is superior to NN. Analysis of performances of NN with different number of neurons or different architectures indicate that the effects of NGN cannot be accounted for an increased number of network elements, but rather they are specifically due to astrocytes. Furthermore, the relative efficacy of NGN vs. NN increases as the complexity of the network increases. These results indicate that artificial astrocytes improve neural network performance, and established the concept of Artificial Neuron-Glia Networks, which represents a novel concept in Artificial Intelligence with implications in computational science as well as in the understanding of brain function.