Proposed Transfer Learning Approach Research Articles

Cross-sensor transfer learning for fire smoke scene detection using variable-bands multi-spectral satellite imagery aided by spectral patterns

ABSTRACT This paper addresses the challenge of training deep learning models for fire smoke scene detection from multi-sensor, multi-spectral satellite imagery, where spectral bands vary and training data is scarce, especially for new sensors. We introduce a novel cross-sensor transfer learning approach enhanced by spectral pattern extraction, notably aided by a deep learning (DL) module called Input Amplification (IA). Our contribution includes the creation of two datasets, Landsat_smk2770 and Sentinel2_smk, with varying spectral bands. We employ a model named IA_VIB_SD which incorporates IA and is designed specifically for fire smoke scene detection. We investigated the transferability of the Landsat_smk2770-trained IA_VIB_SD model to the Sentinel2_smk dataset using various transfer learning techniques, and compared the performance of the transferred model with the benchmark IA_VIB_SD model exclusively trained on pure Sentinel2_smk data. Our proposed transfer learning approach resulted in a transferred model with an average accuracy 5% higher than the benchmark model. Notably, Our proposed transfer learning approach outperformed existing transfer learning methods by more than 1% in terms of the accuracy, even when trained on a mere 10% of the Sentinel2_smk dataset. This research advances fire smoke scene detection from multi-sensor, multi-spectral satellite imagery.

Wearable-based human flow experience recognition enhanced by transfer learning methods using emotion data

Background:Flow experience is a specific positive and affective state that occurs when humans are completely absorbed in an activity and forget everything else. This state can lead to high performance, well-being, and productivity at work. Few studies have been conducted to determine the human flow experience using physiological wearable sensor devices. Other studies rely on self-reported data. Methods:In this article, we use physiological data collected from 25 subjects with multimodal sensing devices, in particular the Empatica E4 wristband, the Emotiv Epoc X electroencephalography (EEG) headset, and the Biosignalplux RespiBAN – in arithmetic and reading tasks to automatically discriminate between flow and non-flow states using feature engineering and deep feature learning approaches. The most meaningful wearable device for flow detection is determined by comparing the performances of each device. We also investigate the connection between emotions and flow by testing transfer learning techniques involving an emotion recognition-related task on the source domain. Results:The EEG sensor modalities yielded the best performances with an accuracy of 64.97%, and a macro Averaged F1 (AF1) score of 64.95%. An accuracy of 73.63% and an AF1 score of 72.70% were obtained after fusing all sensor modalities from all devices. Additionally, our proposed transfer learning approach using emotional arousal classification on the DEAP dataset led to an increase in performances with an accuracy of 75.10% and an AF1 score of 74.92%. Conclusion:The results of this study suggest that effective discrimination between flow and non-flow states is possible with multimodal sensor data. The success of transfer learning using the DEAP emotion dataset as a source domain indicates that emotions and flow are connected, and emotion recognition can be used as a latent task to enhance the performance of flow recognition.

Transfer-learning-based multi-wavelength laser sensor for high fidelity and real-time monitoring of ambient temperature and humidity.

Multi-wavelength laser absorption spectroscopy has the advantages of superior sensitivity, accuracy, and robustness for gas sensing applications, offering an opportunity for the development of high-performance laser-based hygrothermographs. However, accurate and fast determination of gas parameters from multiple spectral features can be quite challenging in the presence of large numbers of features, measurement noise, and increasing demands for real-time measurements. To address this challenge, we propose a transfer-learning-based multi-wavelength laser absorption sensor for the quantitative and simultaneous measurement of temperature and concentration of water vapor, with a focus on real-time monitoring of ambient temperature and relative humidity (RH). A spectral simulation based on the most-updated HITRAN database was employed as the dataset for model pre-training and transfer learning. The experimental dataset was obtained from absorption measurements using a distributed feedback laser that probed multiple water absorption features within the band of 7179-7186c m -1. To evaluate the sensor performance, mean absolute error, error distribution, and linearity were selected. In the presence of an insufficient experimental dataset for direct data training, the proposed transfer learning approach outperformed the traditional deep learning method with a lower prediction error of 0.14°C and 0.42% for temperature and RH, respectively, as compared to the values of 0.84°C and 0.66% obtained using the traditional deep learning method. Finally, the fast data post-processing performance of the proposed transfer learning approach was demonstrated in a field test against the conventional baseline fitting method.

Classification of benign and malignant subtypes of breast cancer histopathology imaging using hybrid CNN-LSTM based transfer learning

BackgroundGrading of cancer histopathology slides requires more pathologists and expert clinicians as well as it is time consuming to look manually into whole-slide images. Hence, an automated classification of histopathological breast cancer sub-type is useful for clinical diagnosis and therapeutic responses. Recent deep learning methods for medical image analysis suggest the utility of automated radiologic imaging classification for relating disease characteristics or diagnosis and patient stratification.MethodsTo develop a hybrid model using the convolutional neural network (CNN) and the long short-term memory recurrent neural network (LSTM RNN) to classify four benign and four malignant breast cancer subtypes. The proposed CNN-LSTM leveraging on ImageNet uses a transfer learning approach in classifying and predicting four subtypes of each. The proposed model was evaluated on the BreakHis dataset comprises 2480 benign and 5429 malignant cancer images acquired at magnifications of 40×, 100×, 200× and 400×.ResultsThe proposed hybrid CNN-LSTM model was compared with the existing CNN models used for breast histopathological image classification such as VGG-16, ResNet50, and Inception models. All the models were built using three different optimizers such as adaptive moment estimator (Adam), root mean square propagation (RMSProp), and stochastic gradient descent (SGD) optimizers by varying numbers of epochs. From the results, we noticed that the Adam optimizer was the best optimizer with maximum accuracy and minimum model loss for both the training and validation sets. The proposed hybrid CNN-LSTM model showed the highest overall accuracy of 99% for binary classification of benign and malignant cancer, and, whereas, 92.5% for multi-class classifier of benign and malignant cancer subtypes, respectively.ConclusionTo conclude, the proposed transfer learning approach outperformed the state-of-the-art machine and deep learning models in classifying benign and malignant cancer subtypes. The proposed method is feasible in classification of other cancers as well as diseases.

Application of Fuzzy Deep Neural Networks for Covid 19 diagnosis through chest Radiographs

Background: The increasing number of COVID-19 patients around the world and the limited number of detection kits pose a challenge in determining the presence of the disease. Imaging modalities such as X-rays are commonly used because they are readily available and cost-effective. Deep learning has proved to be an excellent tool because of the abundance of online medical images in various medical modalities, such as X-Ray, computerized tomography (CT) Scan, and magnetic resonance imaging (MRI). A large number of medical research projects have been proposed and launched since early 2020 due to the overwhelming use of deep learning techniques in medical imaging. Methods: We have used fuzzy logic and deep learning to determine if chest X-ray images belong to people who have pneumonia related to COVID-19 and people who have interstitial pneumonias that aren't related to COVID-19. Results: In comparison to the current literature, the proposed transfer learning approach is more successful. It is possible to classify covid, viral, and bacterial pneumonia or a healthy patient using ResNet 18 Architecture's four-class classifiers. The proposed method achieved a 97% classification accuracy, 96% precision, and 98% recall in the case of COVID-19 detection using chest X-ray images, which demonstrates the importance of deep learning in medical image diagnosis. Furthermore, the results demonstrate that the proposed technique has the maximum sensitivity rate, with 97.1% ratio. Finally, with a 97.47% F1-score rate, the proposed strategy yields the highest value when compared to the others. Conclusions: DeepLearning techniques and fuzzy features resulted in an improved classification ability, with an accuracy rate of up to 97.7% using ResNet 18, which is a better value when compared to the remaining techniques. Classification of COVID-19 scans and other pneumonia cases have been done successfully by demonstrating the potential for applying such deep learning techniques in the near future.

Motion Prediction for Autonomous Vehicle using Deep Learning Architecture and Transfer Learning

In the current scenario, Autonomous Vehicle (AV) technology has become one of the more futuristic approaches in the automotive industry since it aims to enhance driving safety, driving comfort, and its economy, adding to reduced obstacle collision or traffic accident rate. Motion planning contributes a vital part in autonomous driving, serving as a fundamental building block enabling the AV to move further. Though there are several traditional approaches to implement motion planning, yet challenges remain regarding guaranteed performance and safety under all driving circumstances. With the impressive advancement of deep learning technologies, many researchers have tried to develop end-to-end motion planning approaches using deep learning, which generally employ Deep Neural Networks (DNNs) to directly map the raw sensor data (e.g., point clouds and images) to planned trajectories(e.g., yaw velocity and steering angle). However, accurate motion prediction is still critical for autonomous driving where there’s a transfer learning approach which emphasizes on reusing a trained model for various applications. This could possibly improve the accuracy of motion prediction steadily. Considering this in mind, a transfer learning motion prediction approach for autonomous driving employing the DNN is proposed here, also providing visual element representing the predicted path, by streamlining the use of the toolkit available at hand. The output responses are discussed and compared employing the existing prediction model and the proposed model. The experimental results exhibited improved performance of motion prediction of the proposed transfer learning approach proving better accuracy comparitively with the existing Resnet50 model.

Hand Gesture Recognition via Transient sEMG Using Transfer Learning of Dilated Efficient CapsNet: Towards Generalization for Neurorobotics

There has been an accelerated surge in utilizing the deep neural network to decode central and peripheral activations of the human nervous system to boost the spatiotemporal resolution of neural interfaces used in human-centered robotic systems, such as prosthetics, and exoskeletons. Deep learning methods are proven to achieve high accuracy but are also challenged by their assumption of having access to massive training samples. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Objective:</i> In this letter, we propose Dilated Efficient CapsNet to improve the predictive performance when the available individual data is minimal and not enough to train an individualized network for controlling a personalized robotic system. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Method:</i> We proposed the concept of transfer learning for a new design of the dilated efficient capsular neural network to relax the need of having access to massive individual data and utilize the field knowledge which can be learned from a group of participants. In addition, instead of using complete sEMG signals, we only use the transient phase, reducing the volume of training samples to 20% of the original and maximizing the agility. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Results:</i> In experiments, we validate the performance with various amounts of injected personalized training data (from 25% to 100% of transient phase). The results support the use of the proposed transfer learning approach based on the dilated capsular neural network when the knowledge domain learned on a small number of subjects can be utilized to minimize the need for new data from new subjects. The model focuses only on the transient phase which is a challenging neural interfacing problem.

Big Transfer Learning for Fine Art Classification.

Automatic classification and retrieval of fine art collections have received much attention in recent years. In this article, we explore the applicability of convolutional neural networks (CNNs) for art-related image classification tasks. To examine how hyperparameters affect model performance, we use different hyperparameters in our experiments and find that a higher resolution and appropriate training steps with mix-up can improve model performance. To determine how transfer learning affects the final results, we systematically compare the efforts of five weight initializations of the models for different tasks. We show that fine-tuning networks pretrained on a larger dataset have better generalizability. This phenomenon shows the a priori knowledge that models learn in the real world also applies to the art world, and we call this method as big transfer learning (BiT). Through extensive experiments on fine art classification, we demonstrate that the proposed transfer learning approach outperforms the previous work by a large margin and achieves state-of-the-art performance in the art field. Furthermore, to show how computers capture features in paintings to make classifications, we visualized the results of different classification tasks to help us understand the operation mechanism of the models. Additionally, we use our models to retrieve paintings by analyzing different image similarity aspects. The results show that models can be employed to retrieve paintings even if they are computer-generated.

DeepSpectrumLite: A Power-Efficient Transfer Learning Framework for Embedded Speech and Audio Processing From Decentralized Data.

Deep neural speech and audio processing systems have a large number of trainable parameters, a relatively complex architecture, and require a vast amount of training data and computational power. These constraints make it more challenging to integrate such systems into embedded devices and utilize them for real-time, real-world applications. We tackle these limitations by introducing DeepSpectrumLite, an open-source, lightweight transfer learning framework for on-device speech and audio recognition using pre-trained image Convolutional Neural Networks (CNNs). The framework creates and augments Mel spectrogram plots on the fly from raw audio signals which are then used to finetune specific pre-trained CNNs for the target classification task. Subsequently, the whole pipeline can be run in real-time with a mean inference lag of 242.0 ms when a DenseNet121 model is used on a consumer-grade Motorola moto e7 plus smartphone. DeepSpectrumLite operates decentralized, eliminating the need for data upload for further processing. We demonstrate the suitability of the proposed transfer learning approach for embedded audio signal processing by obtaining state-of-the-art results on a set of paralinguistic and general audio tasks, including speech and music emotion recognition, social signal processing, COVID-19 cough and COVID-19 speech analysis, and snore sound classification. We provide an extensive command-line interface for users and developers which is comprehensively documented and publicly available at https://github.com/DeepSpectrum/DeepSpectrumLite.

Learning to Discover Explainable Clinical Features With Minimum Supervision.

PurposeTo compare supervised transfer learning to semisupervised learning for their ability to learn in-depth knowledge with limited data in the optical coherence tomography (OCT) domain.MethodsTransfer learning with EfficientNet-B4 and semisupervised learning with SimCLR are used in this work. The largest public OCT dataset, consisting of 108,312 images and four categories (choroidal neovascularization, diabetic macular edema, drusen, and normal) is used. In addition, two smaller datasets are constructed, containing 31,200 images for the limited version and 4000 for the mini version of the dataset. To illustrate the effectiveness of the developed models, local interpretable model-agnostic explanations and class activation maps are used as explainability techniques.ResultsThe proposed transfer learning approach using the EfficientNet-B4 model trained on the limited dataset achieves an accuracy of 0.976 (95% confidence interval [CI], 0.963, 0.983), sensitivity of 0.973 and specificity of 0.991. The semisupervised based solution with SimCLR using 10% labeled data and the limited dataset performs with an accuracy of 0.946 (95% CI, 0.932, 0.960), sensitivity of 0.941, and specificity of 0.983.ConclusionsSemisupervised learning has a huge potential for datasets that contain both labeled and unlabeled inputs, generally, with a significantly smaller number of labeled samples. The semisupervised based solution provided with merely 10% labeled data achieves very similar performance to the supervised transfer learning that uses 100% labeled samples.Translational RelevanceSemisupervised learning enables building performant models while requiring less expertise effort and time by using to good advantage the abundant amount of available unlabeled data along with the labeled samples.

Efficient generative transfer learning framework for the detection of COVID-19

Deep learning plays a major role in detecting the presence of Coronavirus 2019 (COVID-19) and demands huge data. Availability of annotated data is a hurdle in using Deep learning technique. To enhance the accuracy of detection Deep Convolutional Generative Adversarial Network (DCGAN) is used to generate synthetic data. Densenet-201 is identified as the deep learning framework to detect COVID-19 from X-ray images. In this research, to validate the effectiveness of the Densenet-201, we explored conventional machine learning approaches such as SVM, Random Forest and Convolutional Neural Network (CNN). The feature map for training the machine learning approaches are extracted using Densenet-201 as feature extractor. The results show that Densenet-201 as feature representation with SVM is performing well in detecting COVID-19 with high accuracy. Moreover we experimented the proposed methodology without using DCGAN as well. DenseNet-201 based approach is capable of detecting the presence of COVID-19 with high accuracy. Experiments demonstrated that the proposed transfer learning approach based on DenseNet-201 along with DCGAN based augmentation outperforms the State of the art approaches like ResNet50, CNN, and VGG-16.

Deep learning techniques elucidate and modify the shape factor to extend the effective medium theory beyond its original formulation

The effective medium theories (EMTs) can reliably approximate the property of a composite using properties of the inclusion and matrix phase. However, their inherent assumptions and the availability of mathematical forms for describing the inclusion structure limit their accuracy and applicability. In this work, we utilize the capabilities of a deep learning method to ameliorate the latter restriction for a particular EMT formulation. Our deep learning models elucidate the inclusion structure using several physics-based descriptors and can be easily adapted for other inclusion shapes through transfer learning. Using our models, we shed light on the interpretation of the shape factor in the chosen EMT. More importantly, we extend, not replace, the EMT for cases beyond its original formulation. Our proposed transfer learning approach requires relatively low computation cost and a small sample number, making it especially useful when new data is limited.

Deep learning for COVID-19 detection based on CT images

COVID-19 has tremendously impacted patients and medical systems globally. Computed tomography images can effectively complement the reverse transcription-polymerase chain reaction testing. This study adopted a convolutional neural network for COVID-19 testing. We examined the performance of different pre-trained models on CT testing and identified that larger, out-of-field datasets boost the testing power of the models. This suggests that a priori knowledge of the models from out-of-field training is also applicable to CT images. The proposed transfer learning approach proves to be more successful than the current approaches described in literature. We believe that our approach has achieved the state-of-the-art performance in identification thus far. Based on experiments with randomly sampled training datasets, the results reveal a satisfactory performance by our model. We investigated the relevant visual characteristics of the CT images used by the model; these may assist clinical doctors in manual screening.

Few-shot learning of Parkinson’s disease speech data with optimal convolution sparse kernel transfer learning

The classification of Parkinson’s disease speech data is useful and popular. However, the existing public Parkinson’s disease(PD) speech datasets are characterized by small sample sizes, and the possible reason is that labeled speech data from PD patients are scarce. To solve the few-shot problem, a PD classification algorithm based on sparse kernel transfer learning combined with simultaneous sample and feature selection is proposed in this paper. Sparse kernel transfer learning is used to extract the effective structural information of PD speech features from public datasets as source domain data, and the fast alternating direction method of multipliers (ADMM) iteration is improved to enhance the information extraction performance. First, features are extracted from a public speech dataset to construct a feature dataset as the source domain. Then, the PD target domain, including the training and test datasets, is encoded by convolution sparse coding, which can extract more in-depth information. Next, simultaneous optimization is implemented. To further improve the classification performance, a convolution kernel optimization mechanism is designed. In the experimental section, two representative PD speech datasets are used for verification; the first dataset is a frequently used public dataset, and the second dataset is constructed by the authors. Over ten relevant algorithms are compared with the proposed method. The results show that the proposed algorithm achieves obvious improvements in terms of classification accuracy. The study also found that the improvements are considerable when compared with nontransfer learning approaches, demonstrating that the proposed transfer learning approach is more effective and has an acceptable time cost.

Transfer learning based methodology for migration and application of fault detection and diagnosis between building chillers for improving energy efficiency

Faulty operation of HVAC chillers may result in increased energy usage, reduced thermal comfort, and increased maintenance costs. Establishing reliable fault detection and diagnosis (FDD) methods can effectively avoid these adverse effects. Most existing FDD studies only focus on training specific fault diagnostic models for particular chillers. Very few works have been found to address the issue of how to transfer FDD between different chillers, which is a promising but challenging task. This paper presents a generic framework for transferring the prior knowledge of an information-rich source chiller to build the diagnostic model for a new target chiller. Heterogeneous data standardization is firstly presented to standardize the original datasets of different chillers. Domain adaption transfer learning is then applied to overcome the domain shift introduced by the differences of feature distribution across chillers, and domain adversarial neural network is used to generate the diagnostic model for the target chiller where only easy-to-collect normal operation data of the target chiller and the prior knowledge are used for training. To verify the presented transfer learning, two screw chillers are used as the source and target chillers accordingly, and large quantities of fault simulation experiments are performed on them. Results indicate that the transferred diagnostic model for the target chiller yields decent diagnostic performance, the accuracy for the first and second cycles are 81.27% and 74.93%, respectively. In comparison with five conventional machine learning models with accuracy less than 55%, the proposed transfer learning approach has significant performance advantages and practical application potential.

Improving phoneme recognition of throat microphone speech recordings using transfer learning

Throat microphones (TM) are a type of skin-attached non-acoustic sensors, which are robust to environmental noise but carry a lower signal bandwidth characterization than the traditional close-talk microphones (CM). Attaining high-performance phoneme recognition is a challenging task when the training data from a degrading channel, such as TM, is limited. In this paper, we address this challenge for the TM speech recordings using a transfer learning approach based on the stacked denoising auto-encoders (SDA). The proposed transfer learning approach defines an SDA-based domain adaptation framework to map the source domain CM representations and the target domain TM representations into a common latent space, where the mismatch across TM and CM is eliminated to better train an acoustic model and to improve the TM phoneme recognition. For the phoneme recognition task, we use the convolutional neural network (CNN) and the hidden Markov model (HMM) based CNN/HMM hybrid system, which delivers better acoustic modeling performance compared to the conventional Gaussian mixture model (GMM) based models. In the experimental evaluations, we observed more than 12% relative phoneme error rate (PER) improvement for the TM recordings with the proposed transfer learning approach compared to baseline performances.

Tridirectional Transfer Learning for Predicting Gastric Cancer Morbidity.

Our previous study has constructed a deep learning model for predicting gastrointestinal infection morbidity based on environmental pollutant indicators in some regions in central China. This article aims to adapt the prediction model for three purposes: 1) predicting the morbidity of a different disease in the same region; 2) predicting the morbidity of the same disease in a different region; and 3) predicting the morbidity of a different disease in a different region. We propose a tridirectional transfer learning approach, which achieves the abovementioned three purposes by: 1) developing a combined univariate regression and multivariate Gaussian model for establishing the relationship between the morbidity of the target disease and that of the source disease together with the high-level pollutant features in the current source region; 2) using mapping-based deep transfer learning to extend the current model to predict the morbidity of the source disease in both source and target regions; and 3) applying the pattern of the combined model in the source region to the extended model to derive a new combined model for predicting the morbidity of the target disease in the target region. We select gastric cancer as the target disease and use the proposed transfer learning approach to predict its morbidity in the source region and three target regions. The results show that, given only a limited number of labeled samples, our approach achieves an average prediction accuracy of over 80% in the source region and up to 78% in the target regions, which can contribute considerably to improving medical preparedness and response.

A probabilistic approach for calibration time reduction in hybrid EEG\u2013fTCD brain\u2013computer interfaces

BackgroundGenerally, brain–computer interfaces (BCIs) require calibration before usage to ensure efficient performance. Therefore, each BCI user has to attend a certain number of calibration sessions to be able to use the system. However, such calibration requirements may be difficult to fulfill especially for patients with disabilities. In this paper, we introduce a probabilistic transfer learning approach to reduce the calibration requirements of our EEG–fTCD hybrid BCI designed using motor imagery (MI) and flickering mental rotation (MR)/word generation (WG) paradigms. The proposed approach identifies the top similar datasets from previous BCI users to a small training dataset collected from a current BCI user and uses these datasets to augment the training data of the current BCI user. To achieve such an aim, EEG and fTCD feature vectors of each trial were projected into scalar scores using support vector machines. EEG and fTCD class conditional distributions were learnt separately using the scores of each class. Bhattacharyya distance was used to identify similarities between class conditional distributions obtained using training trials of the current BCI user and those obtained using trials of previous users.ResultsExperimental results showed that the performance obtained using the proposed transfer learning approach outperforms the performance obtained without transfer learning for both MI and flickering MR/WG paradigms. In particular, it was found that the calibration requirements can be reduced by at least 60.43% for the MI paradigm, while at most a reduction of 17.31% can be achieved for the MR/WG paradigm.ConclusionsData collected using the MI paradigm show better generalization across subjects.

Catheter segmentation in X-ray fluoroscopy using synthetic data and transfer learning with light U-nets

Background and objectivesAutomated segmentation and tracking of surgical instruments and catheters under X-ray fluoroscopy hold the potential for enhanced image guidance in catheter-based endovascular procedures. This article presents a novel method for real-time segmentation of catheters and guidewires in 2d X-ray images. We employ Convolutional Neural Networks (CNNs) and propose a transfer learning approach, using synthetic fluoroscopic images, to develop a lightweight version of the U-Net architecture. Our strategy, requiring a small amount of manually annotated data, streamlines the training process and results in a U-Net model, which achieves comparable performance to the state-of-the-art segmentation, with a decreased number of trainable parameters.MethodsThe proposed transfer learning approach exploits high-fidelity synthetic images generated from real fluroscopic backgrounds. We implement a two-stage process, initial end-to-end training and fine-tuning, to develop two versions of our model, using synthetic and phantom fluoroscopic images independently. A small number of manually annotated in-vivo images is employed to fine-tune the deepest 7 layers of the U-Net architecture, producing a network specialized for pixel-wise catheter/guidewire segmentation. The network takes as input a single grayscale image and outputs the segmentation result as a binary mask against the background.ResultsEvaluation is carried out with images from in-vivo fluoroscopic video sequences from six endovascular procedures, with different surgical setups. We validate the effectiveness of developing the U-Net models using synthetic data, in tests where fine-tuning and testing in-vivo takes place both by dividing data from all procedures into independent fine-tuning/testing subsets as well as by using different in-vivo sequences. Accurate catheter/guidewire segmentation (average Dice coefficient of ~ 0.55, ~ 0.26 and ~ 0.17) is obtained with both U-Net models. Compared to the state-of-the-art CNN models, the proposed U-Net achieves comparable performance ( ± 5% average Dice coefficients) in terms of segmentation accuracy, while yielding a 84% reduction of the testing time. This adds flexibility for real-time operation and makes our network adaptable to increased input resolution.ConclusionsThis work presents a new approach in the development of CNN models for pixel-wise segmentation of surgical catheters in X-ray fluoroscopy, exploiting synthetic images and transfer learning. Our methodology reduces the need for manually annotating large volumes of data for training. This represents an important advantage, given that manual pixel-wise annotations is a key bottleneck in developing CNN segmentation models. Combined with a simplified U-Net model, our work yields significant advantages compared to current state-of-the-art solutions.

Personalizing Heart Rate-Based Seizure Detection Using Supervised SVM Transfer Learning.

Objective: Automated seizure detection is a key aspect of wearable seizure warning systems. As a result, the quality of life of refractory epilepsy patients could be improved. Most state-of-the-art algorithms for heart rate-based seizure detection use a so-called patient-independent approach, which do not take into account patient-specific data during algorithm training. Although such systems are easy to use in practice, they lead to many false detections as the ictal heart rate changes are patient-dependent. In practice, only a limited amount of accurately annotated patient data is typically available, which makes it difficult to create fully patient-specific algorithms.Methods: In this context, this study proposes for the first time a new transfer learning approach that allows to personalize heart rate-based seizure detection by using only a couple of days of data per patient. The algorithm was evaluated on 2,172 h of single-lead ECG data from 24 temporal lobe epilepsy patients including 227 focal impaired awareness seizures.Results: The proposed personalized approach resulted in an overall sensitivity of 71% with 1.9 false detections per hour. This is an average decrease in false detection rate of 37% compared to the reference patient-independent algorithm using only a limited amount of personal seizure data. The proposed transfer learning approach adapts faster and more robustly to patient-specific characteristics than other alternatives for personalization in the literature.Conclusion: The proposed method allows an easy implementable solution to personalize heart rate-based seizure detection, which can improve the quality of life of refractory epilepsy patients when used as part of a multimodal seizure detection system.