Ing. Zdeněk Buk, Ph.D.

Článek

10.1088/1748-0221/20/01/C01028

Katedra teoretické informatiky
Katedra aplikované matematiky

Semiconductor hybrid pixel detectors with Timepix 3 chips developed by Medipix collaboration at CERN can simultaneously measure deposited energy and time of arrival of individual particle hits in all 256 × 256 pixels with 55 μm pitch size. Leveraging the single-particle detection sensitivity of these chips, there is a potential to develop algorithms for classifying detected single particles into distinct categories corresponding to different particle types. In this study, various machine learning models are introduced, such as recurrent and feedforward neural networks or gradient boosted decision trees, designed to facilitate the classification of single particle events into distinct classes associated to electrons & photons, alpha particles, heavy nuclei (except alpha particles), low energy protons (E ≲ 100 MeV) and high energy protons (E ≳ 100 MeV). All models achieve outcomes with the true positive rate nearing 100% across all classes. The Gaussian Mixture unsupervised machine learning technique is used to differentiate between electron and photon radiation components. The model effectively distinguished between high-energy electrons and low-energy photons, achieving performance comparable to conventionally used heuristic decision trees. All models are trained and tested on an extensive database of experimental data obtained from controlled radiation source experiments.

Článek

10.1016/j.forsciint.2024.112272

Katedra teoretické informatiky
Katedra počítačových systémů

Objective: Age-at-death estimation is usually done manually by experts. As such, manual estimation is subjective and greatly depends on the past experience and proficiency of the expert. This becomes even more critical if experts need to evaluate individuals with unknown population affinity or with affinity that they are not familiar with. The purpose of this study is to design a novel age-at-death estimation method allowing for automatic evaluation on computers, thus eliminating the human factor. Methods: We used a traditional machine-learning approach with explicit feature extraction. First, we identified and described the features that are relevant for age-at-death estimation. Then, we created a multi-linear regression model combining these features. Finally, we analysed the model performance in terms of Mean Absolute Error (MAE), Mean Bias Error (MBE), Slope of Residuals (SoR) and Root Mean Squared Error (RMSE). Results: The main result of this study is a population-independent method of estimating an individual's age-at- death using the acetabulum of the pelvis. Apart from data acquisition, the whole procedure of pre-processing, feature extraction and age estimation is fully automated and implemented as a computer program. This program is a part of a freely available web-based software tool called CoxAGE3D, which is available at https://co xage3d.fit.cvut.cz/. Based on our dataset, the MAE of the presented method is about 10.7 years. In addition, five population-specific models for Thai, Lithuanian, Portuguese, Greek and Swiss populations are also given. The MAEs for these populations are 9.6, 9.8, 10.8, 10.5 and 9.2 years, respectively. Our age-at-death estimation method is suitable for individuals with unknown population affinity and provides acceptable accuracy. The age estimation error cannot be completely eliminated, because it is a consequence of the variability of the ageing process of different individuals not only across different populations but also within a certain population.

Článek

10.1016/j.forsciint.2023.111765

Katedra teoretické informatiky
Katedra počítačových systémů

This work presents an automated data-mining model for age-at-death estimation based on 3D scans of the auricular surface of the pelvic bone. The study is based on a multi-population sample of 688 individuals (males and females) originating from one Asian and five European identified osteological collections. Our method requires no expert knowledge and achieves similar accuracy compared to traditional subjective methods. Apart from data acquisition, the whole procedure of pre-processing, feature extraction and age estimation is fully automated and implemented as a computer program. This program is a part of a freely available web-based software tool called CoxAGE3D. This software tool is available at https://cox-age3d.fit.cvut.cz/ Our age-at-death estimation method is suitable for use on individuals with known/un-known population affinity and provides moderate correlation between the estimated age and actual age (Pearson's correlation coefficient is 0.56), and a mean absolute error of 12.4 years.& COPY; 2023 Elsevier B.V. All rights reserved.

Stať ve sborníku

10.1109/informatics57926.2022.10083446

Katedra teoretické informatiky

The paper presents an age-at-death estimation model based on artificial neural networks with no explicit feature extraction, thus, completely eliminating the need for expert knowledge. As input information, it uses a 3D surface scan of the acetabulum, and as the output, it provides an estimated age-at-death. This study is based on a heterogeneous multipopulational database composed of 943 adult ossa coxae coming from 380 males and 327 females. The mean absolute error of our model for this database is about 12.4 years. The correlation coefficient between actual and estimated age-at-death is 0.6. This clearly demonstrates that our model captures age-related morphological changes of the shape and surface of the acetabulum.

Článek

10.1038/s41598-022-13983-8

Katedra teoretické informatiky
Katedra počítačových systémů

Age-at-death estimation of adult skeletal remains is a key part of biological profile estimation, yet it remains problematic for several reasons. One of them may be the subjective nature of the evaluation of age-related changes, or the fact that the human eye is unable to detect all the relevant surface changes. We have several aims: (1) to validate already existing computer models for age estimation; (2) to propose our own expert system based on computational approaches to eliminate the factor of subjectivity and to use the full potential of surface changes on an articulation area; and (3) to determine what age range the pubic symphysis is useful for age estimation. A sample of 483 3D representations of the pubic symphyseal surfaces from the ossa coxae of adult individuals coming from four European (two from Portugal, one from Switzerland and Greece) and one Asian (Thailand) identified skeletal collections was used. A validation of published algorithms showed very high error in our dataset-the Mean Absolute Error (MAE) ranged from 16.2 and 25.1 years. Two completely new approaches were proposed in this paper: SASS (Simple Automated Symphyseal Surface-based) and AANNESS (Advanced Automated Neural Network-grounded Extended Symphyseal Surface-based), whose MAE values are 11.7 and 10.6 years, respectively. Lastly, it was demonstrated that our models could estimate the age-at-death using the pubic symphysis over the entire adult age range. The proposed models offer objective age estimates with low estimation error (compared to traditional visual methods) and are able to estimate age using the pubic symphysis across the entire adult age range.

Stať ve sborníku

Katedra teoretické informatiky
Katedra počítačových systémů
Katedra aplikované matematiky

The paper introduces a novel age-at-death estimation model based on Convolutional Neural Network (CNN). The model uses 3D scan of human pubic symphysis as an input and estimates the age-at-death of the individual as an output. The Mean Absolute Error (MAE) of this model is about 10.6 years for individuals between 18 and 92 years of age-at-death. Moreover, the results of the study indicate that pubic symphysis can be used to estimate the age of individuals across the entire age range. The study involved a sample of 483 bone scans collected from 374 individuals (from which 109 individuals provided both left and right pubic symphysis).

Článek

10.1016/j.forsciint.2018.03.047

Katedra teoretické informatiky
Katedra počítačových systémů

The assessment of age-at-death is an important and challenging part of investigations of human skeletal remains. The main objective of the present study was to apply different mathematical approaches in order to reach more accurate and reliable results in age estimation. A multi-ethnic dataset (n = 941) of evaluated age-related changes on the pubic symphysis and the auricular surface of the hip bone was used. Two research groups examined nine different mathematical approaches. The best results were reached by Multi-linear regression, followed by the Collapsed regression model, with MAE values of 9.7 and 9.9 years, respectively, and with RMSE values of 12.1 and 12.2, respectively. The mean accuracy of decision tree models ranged between 30.7% and 72.3%, with the model using only the PUSx indicator performing the best. Moreover, our results indicate that the limiting factor of age estimation can be the visual evaluation of age-related changes. Further research is required to objectify the proposed methods for estimating age. (c) 2018 Elsevier B.V. All rights reserved.

Stať ve sborníku

Katedra teoretické informatiky
Katedra počítačových systémů

This paper compares traditional averages-based model with other various age estimation models in the range from the simplest to the advanced ones, and introduces novel Tabular Constrained Multiple-linear Regression (TCMLR) model. This TCMLR model has similar complexity as traditional averages-based model (it can by evaluated manually), but improves the mean absolute error in average about 0.30 years (approx. 3.6 months) for males, and 0.18 years (approx. 2.2 months) for females, respectively. For all models, the chronological age of an individual is estimated from mineralization stages of dentition. This study was based on a sample of 976 orthopantomographs taken of 662 boys and 314 girls of Czech nationality aged between 2.7 and 20.5 years.

Článek

10.1016/j.forsciint.2017.08.005

Katedra teoretické informatiky
Katedra počítačových systémů

In order to analyze and improve the dental age estimation in children and adolescents for forensic purposes, 22 age estimation methods were compared to a sample of 976 orthopantomographs (662 males, 314 females) of healthy Czech children and adolescents aged between 2.7 and 20.5 years. All methods are compared in terms of the accuracy and complexity and are based on various data mining methods or on simple mathematical operations. The winning method is presented in detail. The comparison showed that only three methods provide the best accuracy while remaining user-friendly. These methods were used to build a tabular multiple linear regression model, an M5P tree model and support vector machine model with first-order polynomial kernel. All of them have mean absolute error (MAE) under 0.7 years for both males and females. The other well-performing data mining methods (RBF neural network, K-nearest neighbors, Kstar, etc.) have similar or slightly better accuracy, but they are not user-friendly as they require computing equipment and the implementation as computer program. The lowest estimation accuracy provides the traditional model based on age averages (MAE under 0.96 years). Different relevancy of various teeth for the age estimation was found. This finding also explains the lowest accuracy of the traditional averages-based model. In this paper, a technique for missing data replacement for the cases with missing teeth is presented in detail as well as the constrained tabular multiple regression model. Also, we provide free age prediction software based on this wining model.

Stať ve sborníku

Katedra teoretické informatiky

The paper presents application of OpenCLLink in Wolfram Mathematica to accelerate fully recurrent neural networks using GPU. We also show the idea of automatically generated parts of source code using SymbolicC.