Multiple Layer Perception Research Articles

Efficient identification of Alzheimer’s brain dynamics with Spatial-Temporal Autoencoder: A deep learning approach for diagnosing brain disorders

Alzheimer’s disease (AD) is a progressive neurological disorder seriously affecting cognitive and behavior abilities of the older people. Accurate and early diagnosis of AD is critical for improving the therapeutic effect and alleviate the clinical symptom. In this work, we proposed an automatic EEG-based diagnosis scheme for AD patients with deep learning methods. A Spatial-Temporal Autoencoder (STAE) framework with Convolutional Neural Network (CNN)- Long Short-Term-Memory (LSTM) generative model was designed to estimate latent factors of the observed oscillatory activity in the brain from multi-channels electroencephalogram (EEG) signals via unsupervised learning. Based on latent factor analysis, Alzheimer’s brain dynamics on single-trials were inferred and temporal evolution of latent brain state was analyzed in low-dimensional state space. The study mainly showed that: i) the brain state trajectories of AD patients were distinct from healthy subjects, resulting in different forms of ring manifolds and allowing to accurately identify AD; ii) experimental results demonstrated the efficiency and flexibility of the proposed deep learning-based diagnosis scheme, by which the classification of AD patients and the normal based on clinical EEG dataset achieved an accuracy of 96.30%, a sensitivity of 97.73% and a specificity of 94.69% with a multiple layer perception (MLP) classifier; iii) compared with other approaches for latent brain dynamics estimation, STAE exhibited superior performance with high accuracy of AD identification and strongly robust against instabilities in EEG recordings. The present results reveal the neuropathological mechanism of Alzheimer’s disease with brain dynamics variations and provide a feasible diagnosis tool for brain disorders.

Application and validation of internally cured concrete strength characteristics by machine learning

Water is a precious resource, and the demand for potable water is high in many regions globally. Potable water is being used in the construction of buildings, and a substantial amount of water is required for curing concrete elements, which can be saved with modern construction practices like internal or self-curing concrete. The properties of concrete are usually identified through physical testing. Experiments become time-consuming, laborious and costly if multiple parameters in small proportions are to be modified and the results are to be predicted. Predicting nominal mix proportions becomes difficult; alternatively, utilising computational and predicting any specific or multiple outputs by analysing multiple input parameters with randomised data is feasible through machine learning. Considering this, an attempt is made to develop self-curing concrete of M30 grade utilising fractional replacement of 30% coarse aggregates with fly-ash aggregate as an internal curing agent. Different algorithms those that are processed in machine learning deal with prediction and validation of concrete characteristic strength parameters; 1) decision tree regression; 2) random forest regression, and 3) multiple layer perception regression analysis. In this study, the comparison found that the utmost successful machine learning strategy for prediction of concrete strength properties is the decision tree regressor method, with the least error fitting the regression curve.

SPD matrices representing artery anatomy for first-pass effect prediction by aggregated networks with multi-scale attentions

ABSTRACT First-pass Effect(FPE) for Endovascular Therapy(EVT) is associated with good clinical outcome (mRS2) of patients with Acute Ischaemic Stroke (AIS). A rapid and accurate prediction of FPE before EVT can help neurointerventionists plan the procedure and avoid delays in restoration of cerebral blood flow. However, there are rare studies focused on FPE prediction on arterial vessel anatomy immediately. The intractable difficulty lies in extracting discriminative features to represent a wide variety of vessels with irregular vessel shapes. In this paper, we propose a Symmetric Positive Definite(SPD) matrix-based feature representation extracted from the centerline of arterial vessels, encoding the global discriminative information over the artery for predicting FPE. Subsequently, a collaborative network of multi-scale Convolutional Neural Network(CNN) and Multiple Layer Perception (MLP) with specific attentions is developed. Specifically, the CNN is used for capturing the features among the multi-scale local neighbours of the curve. MLPs are used for capturing more prominent global discriminative features at different scales. The attention mechanism is used to better filter and extract the useful information for feature fusion. Quantitative experimental results demonstrate that our proposed method is able to predict FPE accurately, outperforming the manually defined features and traditional machine learning-based methods in this task, regarding the metrics of AUC, precision, sensitivity, specificity and accuracy.

Machine-learning models for prediction of sepsis patients mortality

ObjectivesSepsis is an infection-caused syndrome, that leads to life-threatening organ damage. We aim to develop machine learning models with large-scale data to predict sepsis patients’ mortality. Designwe extracted sepsis patients from two databases, Medical Information Mart for Intensive Care IV (MIMIC-IV) as a train set and Philips eICU Collaborative Research Database as a test set. SettingICUs in multicenter hospitals in the USA during 2012–2019. Patients or participantsA total of 21,680 sepsis-3 patients are included in the study, in which, 3771 patients were dead and 17,909 survived during hospitalization, respectively. InterventionsNo interventions. Main variables of interestBasic information, examination items during hospitalization and some medication and treatment information are incorporated into analyzed. Seven different models were built with a Support vector machine, Decision Tree Classifier, Random Forest, Gradients Boosting, Multiple Layer Perception, Xgboost, light Gradients Boosting to predict dead or live during hospitalization. ResultsAlgorithms with an AUC value in the test set of the top three: light GBM, GBM, Xgboost. Considering the performance of the training set and the test set, the light GBM model performs best, and then the parameters of the model were adjusted, after that the AUC value was 0.99 in the train set, 0.96 in the test set, respectively. ConclusionsModels built with light GBM algorithm from real-world sepsis patients from electronic health records accurately predict whether sepsis patients are dead and can be incorporated into clinical decision tools to enhance the prognosis of the patient and prevent adverse outcomes.

Machine learning for early prediction of sepsis-associated acute brain injury.

BackgroundSepsis-associated encephalopathy (SAE) is defined as diffuse brain dysfunction associated with sepsis and leads to a high mortality rate. We aimed to develop and validate an optimal machine-learning model based on clinical features for early predicting sepsis-associated acute brain injury.MethodsWe analyzed adult patients with sepsis from the Medical Information Mart for Intensive Care (MIMIC III) clinical database. Candidate models were trained using random forest, support vector machine (SVM), decision tree classifier, gradients boosting machine (GBM), multiple layer perception (MLP), extreme gradient boosting (XGBoost), light gradients boosting machine (LGBM) and a conventional logistic regression model. These methods were applied to develop and validate the optimal model based on its accuracy and area under curve (AUC).ResultsIn total, 12,460 patients with sepsis met inclusion criteria, and 6,284 (50.4%) patients suffered from sepsis-associated acute brain injury. Compared other models, the LGBM model achieved the best performance. The AUC for both train set and test set indicated excellent validity (Trainset AUC 0.91, Testset AUC 0.87). Feature importance analysis showed that glucose, age, mean arterial pressure, heart rate, hemoglobin, and length of ICU stay were the top 6 important clinical factors to predict occurrence of sepsis-associated acute brain injury.ConclusionAlmost half of patients admitted to ICU with sepsis had sepsis-associated acute brain injury. The LGBM model better identify patients with sepsis-associated acute brain injury than did other machine-learning models. Glucose, age, and mean arterial pressure were the three most important clinical factors to predict occurrence of sepsis-associated acute brain injury.

STGM: Vehicle Trajectory Prediction Based on Generative Model for Spatial-Temporal Features

For safe and efficient navigation in driving environment, autonomous vehicles are supposed to predict future trajectories of surrounding vehicles dynamically and tackle the uncertainty of environment. Early works view uncertainty as distributions in different modals, and perform this task by enumerating a mass of trajectories, but can hardly cover all the possible modal distributions. Recently, generative model has risen to hold promise for modelling such distributions, something that we will build upon here. Target at predefined modals of vehicle agents, this paper proposes Spatial-Temporal Generative Model (STGM), a new learning algorithm that leverages a stochastic generative model to mitigate modal distributions problem. To be more specific, a Conditional Variational Auto-Encoder (CVAE) is leveraged to construct the framework of STGM, in which the input is augmented into modal distributions to gain modality-wise trajectories. Besides, for the stability of prediction, we also come up with a new modal-wise sampling trick as an alternative of traditional random sampling in CVAE. Considering that multiple layer perception (MLP) based CVAE is incapable of extracting spatial and temporal features effectively, we use two respective encoders before CVAE: Spatial Encoder and Temporal Encoder, and align them with a closed modal set. Additionally, inspired by the Total Probability Formula, we adopt a Modal Prediction Model to refine the confidences of modal-wise trajectories. After empirical evaluation on two public datasets, we find that STGM almost outperforms the baselines, such as Covernet, MTP, and so on.

Dynamic model development for vehicle air conditioners based on physics-guided deep learning

Air conditioners (AC) are responsible for the largest portion of energy use among all auxiliary components of vehicles. Most white-box models developed for dynamic modeling of vapor compression systems (VCS) are not adequate to model vehicle ACs due to stronger dynamics and disturbances caused by frequent door open/close, random numbers of passengers entering/leaving the vehicles as well as intermittent shading effect on roads. In this study, a novel physics-guided deep learning method is proposed for dynamic modeling of vehicle ACs based on both domain knowledge and historical operational data. To maximize the practical values of the model in control and diagnosis of ACs, this research aims at developing an integrated VCS model consisting of individual models of major AC components rather than a black-box model of the entire system. Domain knowledge guides the determination of model inputs and outputs, design of the model structure, and understanding of temporal relationship in developing dynamic heat exchanger models. A newly developed NARX-LSTM-MLP neural network is proposed for heat exchange process modelling. The compressor is modelled by a multiple-layer perception (MLP). Component models are integrated by referring to the physical system structure. Field AC operation data from a city bus collected by IoT sensors are used in this study. Validation results indicate good accordance between measurements and simulation results. The model developed is less expensive, more convenient, and more feasible for health monitoring of numerous vehicle ACs at city level in the context of IoT.

AutoGenome: An AutoML tool for genomic research

Deep learning has achieved great successes in traditional fields like computer vision (CV), natural language processing (NLP), speech processing, and more. These advancements have greatly inspired researchers in genomics and made deep learning in genomics an exciting and popular topic. The convolutional neural network (CNN) and recurrent neural network (RNN) are frequently used to solve genomic sequencing and prediction problems, and multiple layer perception (MLP) and auto-encoders (AE) are frequently used for genomic profiling data like RNA expression data and gene mutation data. Here, we introduce a new neural network architecture-the residual fully-connected neural network (RFCN)-and describe its advantage in modeling genomic profiling data. We also incorporate AutoML algorithms and implement AutoGenome, an end-to-end, automated deep learning framework for genomic studies. By utilizing the proposed RFCN architecture, automatic hyper-parameter search, and neural architecture search algorithms, AutoGenome can automatically train high-performance deep learning models for various kinds of genomic profiling data. To help researchers better understand the trained models, AutoGenome can assess the importance of different features and export the most critical features for supervised learning tasks and the representative latent vectors for unsupervised learning tasks. We expect AutoGenome will become a popular tool in genomic studies.

Attention-based dynamic user modeling and Deep Collaborative filtering recommendation

Deep learning (DL) techniques have been widely used in recommender systems for user modeling and matching function learning based on historical interaction matrix. However, existing DL-based recommendation methods usually perform static user preference modeling by using historical interacted items of the user. In this article, we present a time-aware deep CF framework which contains two stages: dynamic user preference modeling based on attention mechanism and matching score prediction based on DL. In the first stage, short-term user preferences are modeled by the time-aware attention mechanism that fully considered the predicted item, the recent interacted items and their interaction time. The resulting short-term preferences are combined with long-term preferences for dynamic user preference modeling. In the second stage, high-order user-item feature interactions are learned by two types of DL models, Deep Matrix Factorization (DMF) and Multiple-Layer Perception (MLP), and the feature interaction vectors of the two models are fused in the last layer of the model to predict the matching score. Extensive experiments on five datasets indicate that our method is superior to the existing time-aware and DL-based recommendation methods in top-k recommendations significantly and consistently.

Predicting hourly heating load in a district heating system based on a hybrid CNN-LSTM model

Heat loads change dynamically with meteorological conditions and user demand, and the related accurate prediction algorithms are conducive to the realization of optimized automatic control and green heating. However, a traditional load forecasting algorithm exhibits low accuracy and is sensitive to data noise. Hence, it is difficult to satisfy the requirements of industrial applications via forecasting algorithms. The progress in big data technology leads to the construction of a smart district heating system (SDHS) with the next generation technology wherein artificial intelligence algorithms play the role of a chemical catalyst. The operation monitoring of heat exchange stations installed with various Internet of Things (IoT) sensors provides abundant data resources for heating load forecasting based on machine learning algorithms. A novel estimation model based on spatiotemporal hybrid convolution neural network long–short term memory (CNN–LSTM) is presented to predict the heating load more reasonably for the SDHS. This hybrid model deeply integrates the excellent feature extraction ability of CNN and LSTM in spatiotemporal two-dimensional feature space. Furthermore, CNN can extract the heating load's spatial characteristics and influencing factors, and LSTM can extract the heating load's temporal characteristics. Therefore, the CNN–LSTM algorithm can describe the complex change trend of the heating load and lead to a more accurate dynamic model of heating load with high nonlinearity and considerable thermal inertia delay. A detailed performance comparison was conducted between the CNN–LSTM and support vector machine (SVM) and ensemble learning algorithms, such as extra tree regression (ETR), random forest regression (RFR), gradient boosting regression (GBR), multiple layer perception (MLP). The experimental results showed the evident accuracy advantages of prediction performance based on the proposed CNN–LSTM algorithm. The MAPE evaluation indexes of the four heat exchange stations are distributed between 3.1% and 4.1%. Besides, this algorithm exhibits good adaptability for heating load data with different numerical ranges, which can satisfy the field engineering applications' requirements in a more enhanced manner.

Device-free single-user activity recognition using diversified deep ensemble learning

WiFi-based human activity recognition (HAR) aims to recognize human activities in an off-the-shelf manner that only relies on the commercial Wi-Fi devices already installed in environments. The recent trend in HAR research is to train classifiers on top of statistical or deep neural features extracted from channel state information (CSI) data. Unfortunately, existing methods only take into account the temporal-correlation within each CSI subcarrier, while ignoring the spatial-correlation between different subcarriers. This issue has not been fully exploited yet, resulting a limited performance. To address this issue, we propose WiAReS, a WiFi-based device-free activity recognition system that takes both temporal-correlation and spatial-correlation into account. WiAReS embarks on diversified deep ensemble methods 2̌for single-user activity recognition where one user performs a single activity at a given time. More specifically, it adopts convolutional neural network (CNN) to automatically extract features from CSI measurements with the preservation of the locality of both spatial patterns and temporal patterns. To further improve recognition accuracy upon CNN-extracted features, we propose a novel ensemble architecture that fuses a multiple layer perception (MLP), a random forest (RF) and a support vector machine (SVM). Our system obtains the CSI data in PHY layer of off-the-shelf WiFi devices by installing Atheros-CSI-Tool on AR9590 based WiFi network interface cards (NICs). Comprehensive experiments have been conducted in three real environments with environmental variation to evaluate the performance of the proposed WiAReS. The experimental results demonstrate that the proposed WiARes system significantly outperforms existing methods.

Electronic band structure phase diagram of 3D carbon allotropes from machine learning

The unique electronic configuration endows carbon with super-flexible bonding ability, displaying metallic, semi-conducting and insulating features with unprecedented applications. Inspired by the pressure–temperature phase diagram that clearly shows the phases (solid/liquid/gas) of a substance in different conditions, for the first time, we have derived the electronic ‘phase diagram’ using machine learning that can discover complex rules and invisible relationships among multi-variables. Based on SACADA database with 522 three-dimensional carbon allotropes, electronic band gap is studied by using support vector machine, decision tree and multiple-layer perception algorithms. It is identified that density and bond angle are two key factors in determining the electronic phase diagram for distinguishing metallic, semiconducting, and insulating carbon phases, where density relates to bond length and coordination number, while bond angle relates to orbital orientations, both together determine the overlap of wave functions between different orbitals.

Deep Learning-Based Decoding of Constrained Sequence Codes

Constrained sequence (CS) codes, including fixed-length CS codes and variable-length CS codes, have been widely used in modern wireless communication and data storage systems. Sequences encoded with constrained sequence codes satisfy constraints imposed by the physical channel to enable efficient and reliable transmission of coded symbols. In this paper, we propose using deep learning approaches to decode fixed-length and variable-length CS codes. Traditional encoding and decoding of fixed-length CS codes rely on look-up tables (LUTs), which is prone to errors that occur during transmission. We introduce fixed-length constrained sequence decoding based on multiple layer perception (MLP) networks and convolutional neural networks (CNNs), and demonstrate that we are able to achieve low bit error rates that are close to maximum a posteriori probability (MAP) decoding as well as improve the system throughput. Further, implementation of capacity-achieving fixed-length codes, where the complexity is prohibitively high with LUT decoding, becomes practical with deep learning-based decoding. We then consider CNN-aided decoding of variable-length CS codes. Different from conventional decoding where the received sequence is processed bit-by-bit, we propose using CNNs to perform one-shot batch-processing of variable-length CS codes such that an entire batch is decoded at once, which improves the system throughput. Moreover, since the CNNs can exploit global information with batch-processing instead of only making use of local information as in conventional bit-by-bit processing, the error rates can be reduced. We present simulation results that show excellent performance with both fixed-length and variable-length CS codes that are used in the frontiers of wireless communication systems.

Groundwater aquifer potential modeling using an ensemble multi-adoptive boosting logistic regression technique

Machine learning and data-driven models have achieved a favorable reputation in the field of advanced geospatial modeling, particularly for models of groundwater aquifer potential over large areas. Such models built using standalone machine learning techniques retain some uncertainty, including errors associated with the modeling process, sampling approach, and input hyper-parameters. Some of these techniques cannot be applied in data-scarce regions because high bias and variance can lead to oversimplification. Therefore, in the current study, we developed and validated a novel ensemble multi-adaptive boosting logistic regression (MABLR) model for groundwater aquifer potential mapping. This model was validated in a large area of the Gyeongsangbuk-do basin in South Korea and the results were compared to those of different types of machine learning models including multiple-layer perception (MPL), logistic regression (LR), and support vector machine (SVM) models. A forward stepwise LR technique was implemented to assess the importance of contributing morphological factors; we found 15 factors that contributed significantly: topographic wetness index (TWI), topographic roughness index (TRI), stream power index (SPI), topographic position index (TPI), multi-resolution valley bottom flatness (MVBF), slope, aspect, slope length (LS), distance from the river, distance from the fault, profile curvature, plane curvature, altitude, land use/land cover (LULC), and geology. We optimized the MABLR model using a fuzzy logic supervised (FLS) approach with 184 iterations and then validated the results using accuracy assessment metrics including the κ coefficient, root-mean-square error (RMSE), receiver operating characteristics (ROC), and the precision-recall curve (PRC). Our model had superior predictive performance among the models tested, with higher overall goodness-of-fit and validation values according to the κ coefficient (0.819 and 0.781, respectively), ROC (0.917 and 0.838), and PRC (0.931 and 0.872). Our experimental results demonstrate that MABLR is more effective at reducing bias and variance error than other constituent machine learning methods.

Application of Artificial Neural Networks (ANN) Coupled with Near-InfraRed(NIR) Spectroscopy for Detection of Adulteration in Honey

Honey is a naturally sweet and viscous product for which the addition of any substance is prohibited by international regulation. Detection of adulteration in honey is a technical problem: adulteration of honey with invert sugar and syrup may not be reliably detected by direct sugar analysis because its constituents are identical to the major natural components of the honey. Therefore, it is important to develop a rapid and reliable analytical method to detect such additions. We used near-infrared spectroscopy (NIR) combined with principle component analysis (PCA) and artificial neural networks (ANN) modelling to discriminate between honey and corn syrup in adulterated honey. Fifteen honey samples from north-west Croatia (Krapina-Zagorje County) were intentionally supplemented with differing proportions of corn syrup ranging from 10-90%. We collected a total of 460 NIR spectra using the Control Development NIR128L-1.7 spectrophotometer (Control Development, South Bend, Indiana, USA) with their software Spec32 software anda HL-2000 halogen light source. For each of the prepared samples, we measured water content by refractometer (Brouwland, Belgium), conductivity byconductometer (SevenCompact, MettlerToledo, Switzerland), and colour using a PCE-CSM3 colorimeter (PCE Instruments, Germany). Prior to ANN modelling, PCA was used to identify patterns and highlight similarities and differences in data of the individual set of the experiment. The goal of PCA is to extract important information from the data table and to express this information as a set of new orthogonal variables called principal components or factors (PCs or Fs). We conducted PCA of raw spectra using the Unscrambler® X 10.4 software (CAMO software, Norway). Data were divided into ANN model training, test, and validation datasets at a 70:15:15 ratio using the first five PCs. ANNs were calibrated using model training data, and evaluated using model test and model validation datasets for their ability to predict: i) the amount of added adultering substance in honey, ii) water content, iii) conductivity and iv) colour of the adulterated honey. Multiple layer perception (MLP) networks were developed in Statistica v.10.0 software (StatSoft, Tulsa, USA). Back error propagation algorithm available in Statistica v.10.0 was applied for the model training. Model performance was evaluated using R2 and root mean squared error (RMSE) values for model training, test, and validation datasets. Results show that network MLP 5-8-6 with five neurons in the input layer, 8 neurons in the hidden layer and 6 neurons in the output layer predicts the analysed output variables with high precision (R2validation,concentration = 0.995, R2validation,water content = 0.993, R2validation,conductivity = 0.992, R2validation,L = 0.939, R2validation,a = 0.895, R2validation,b = 0.924).

Enhanced Evolutionary Sequential Minimal Optimization Model for Inflation Prediction

The control of inflation rate is at the core of monetary policy making. Therefore, there is very great interest in reliable inflation forecasts by central bankers to help them achieve this aim. The aim of this investigation has been to forecast inflation in case of the United States as accurately as possible. This paper proposes a new forecasting model called Sequential Minimal Organization (SMOreg-3passes) for regression predictions. SMOreg-3passes consists of four steps, they are technical indicators generation, feature selection, normalization regression and regression forecaster. The proposed model evaluated using two regression measurements (Mean Absolute Error (MAE) and Root Mean Square Error (RMSE)). Our evidence from the SMOreg-3passes model suggests that the chronology of time series has great influence on future forecasting and the error in forecasting the past has an exponential impact on the current data. The results showed that the proposed model outperformed the traditional SMO and Multiple Layer Perception (MLP) methods.

How to estimate and predict the expenses incurred by diabetes treatment using Artificial Neural Network (ANN)

Diabetes is considered a great health problem due to its economic importance and the fact that it is a chronic disease. The aim of this study was to determine the costs imposed on diabetic patients using Artificial Neural Network. The study data were gathered by random- ly investigating 396 individuals who referred to Kermanshah diabetic center. In this study, Artificial Neural Network using Multiple Layer Perception (MLP) was used to investigate the costs imposed on diabetic patients. In this research, the variable related to treatment of diabetes was calculated through neural network in 8 different output layers. The eight output layers of expenses included physician's vis- it, medication, tests, hospitalization, radiology, treatment of symptoms, transportation, and counseling. In this study, the highest annual expenses were related to medication, tests, and radiology. In addition, the patients who presented the symptoms spent more money for treatment. Considering need to efficiently use medical facilities, it is necessary use ways which be applied to predict the medical ex- penses. We came to conclusion that neural network had great advantages over the regression model and could be used as an efficient tool for prediction of expenses and can replace the classical and statistical models.

Selection of Air Force Pilot Candidates: A Case Study on the Predictive Accuracy of Discriminant Analysis, Logistic Regression, and Four Neural Network Types

We evaluated the predictive classification accuracy of discriminant analysis, logistic regression and four neural network typologies (multiple layer perceptrons, radial basis networks, probabilistic neural networks, and linear neural networks) on a flight screening program with a pass–fail criterion using several psychometric tests as predictors. A stepwise (for logistic regression and discriminant analysis) and sensitivity (for neural networks) selection procedure identified spatial visualization, eye–hand–foot coordination, and concentration capacity as significant predictors. Performance on the first few flights of the screening program was also retained as a significant predictor of final score. Regarding the accuracy of predictions, logistic regression showed the highest accuracy (77%), with high sensitivity (92%) but low specificity (31%). Discriminant analysis had high sensitivity (77%) and high specificity (64%). However, it had the second lowest accuracy (74%). The best performing neural network type was the multiple layer perception, which showed high sensitivity (85%), the second highest specificity (47%), and high accuracy (76%). Radial basis networks and probabilistic networks both fail to predict correctly the candidates who fail on the flight screening program (0% specificity).

The potential role of dynamic thermal analysis in breast cancer detection.

BackgroundIt is presently well accepted that the breast exhibits a circadian rhythm reflective of its physiology. There is increasing evidence that rhythms associated with malignant cells proliferation are largely non-circadian. Cancer development appears to generate its own thermal signatures and the complexity of these signatures may be a reflection of its degree of development. The limitations of mammography as a screening modality especially in young women with dense breasts necessitated the development of novel and more effective screening strategies with a high sensitivity and specificity. The aim of this prospective study was to evaluate the feasibility of dynamic thermal analysis (DTA) as a potential breast cancer screening tool.Methods173 women undergoing mammography as part of clinical assessment of their breast symptoms were recruited prior to having a biopsy. Thermal data from the breast surface were collected every five minutes for a period of 48 hours using eight thermal sensors placed on each breast surface [First Warning System (FWS), Lifeline Biotechnologies, Florida, USA]. Thermal data were recorded by microprocessors during the test period and analysed using specially developed statistical software. Temperature points from each contra-lateral sensor are plotted against each other to form a thermal motion picture of a lesion's physiological activity. DTA interpretations [positive (abnormal thermal signature) and negative (normal thermal signature)] were compared with mammography and final histology findings.Results118 (68%) of participating patients, were found to have breast cancer on final histology. Mammography was diagnostic of malignancy (M5) in 55 (47%), indeterminate (M3, M4) in 54 (46%) and normal/benign (M1, M2) in 9 (8%) patients. DTA data was available on 160 (92.5%) participants. Using our initial algorithm, DTA was interpreted as positive in 113 patients and negative in 47 patients. Abnormal thermal signatures were found in 76 (72%) out of 105 breast cancer patients and 37 of the 55 benign cases. Then we developed a new algorithm using multiple-layer perception and SoftMax output artificial neural networks (ANN) on a subgroup (n = 38) of recorded files. The sensitivity improved to 76% (16/21) and false positives decreased to 26% (7/27)ConclusionDTA of the breast is a feasible, non invasive approach that seems to be sensitive for the detection of breast cancer. However, the test has a limited specificity that can be improved further using ANN. Prospective multi-centre trials are required to validate this promising modality as an adjunct to screening mammography especially in young women with dense breasts.