Automatic Staging Research Articles

A deep learning based experimental framework for automatic staging of pressure ulcers from thermal images

ABSTRACT Pressure ulcers can be life-threatening, and accurate staging is crucial for proper treatment. Manual methods for ulcer detection and staging are subjective and prone to errors. Since the wound temperature holds valuable and informative data, we propose thermal imaging as a tool to detect the stages of pressure ulcers automatically. This paper implements a deep learning-based framework for the automatic detection and classification of pressure ulcer stages 2–4 using thermal images. 450 thermal images were used to train the classification model. The framework combines ‘R50-FPN’, retrained using transfer learning, for ulcer detection and cropping. An ensemble model is used for ulcer stage classification. The model was trained on a dataset consisting of 68 original images and their augmentations. The proposed method achieves a mAP for object detection of 0.70 and an ulcer stage classification accuracy of 0.9545 during evaluation. Class activation maps were also employed to study the model performance further, indicating that the trained model focuses on the wound periphery and the wound bed for classification. The results show a significant improvement in classification accuracy when compared with related works demonstrating the potential of deep learning-based thermal imaging as an objective method for the automated staging of pressure ulcers.

An automated sleep staging tool based on simple statistical features of mice electroencephalography (EEG) and electromyography (EMG) data.

Electroencephalogram (EEG) and electromyogram (EMG) are fundamental tools in sleep research. However, investigations into the statistical properties of rodent EEG/EMG signals in the sleep-wake cycle have been limited. The lack of standard criteria in defining sleep stages forces researchers to rely on human expertise to inspect EEG/EMG. The recent increasing demand for analysing large-scale and long-term data has been overwhelming the capabilities of human experts. In this study, we explored the statistical features of EEG signals in the sleep-wake cycle. We found that the normalized EEG power density profile changes its lower and higher frequency powers to a comparable degree in the opposite direction, pivoting around 20-30 Hz between the NREM sleep and the active brain state. We also found that REM sleep has a normalized EEG power density profile that overlaps with wakefulness and a characteristic reduction in the EMG signal. Based on these observations, we proposed three simple statistical features that could span a 3D space. Each sleep-wake stage formed a separate cluster close to a normal distribution in the 3D space. Notably, the suggested features are a natural extension of the conventional definition, making it useful for experts to intuitively interpret the EEG/EMG signal alterations caused by genetic mutations or experimental treatments. In addition, we developed an unsupervised automatic staging algorithm based on these features. The developed algorithm is a valuable tool for expediting the quantitative evaluation of EEG/EMG signals so that researchers can utilize the recent high-throughput genetic or pharmacological methods for sleep research.

Breathe out the Secret of the Lung: Video Classification of Exhaled Flows from Normal and Asthmatic Lung Models Using CNN-Long Short-Term Memory Networks

In this study, we present a novel approach to differentiate normal and diseased lungs based on exhaled flows from 3D-printed lung models simulating normal and asthmatic conditions. By leveraging the sequential learning capacity of the Long Short-Term Memory (LSTM) network and the automatic feature extraction of convolutional neural networks (CNN), we evaluated the feasibility of the automatic detection and staging of asthmatic airway constrictions. Two asthmatic lung models (D1, D2) with increasing levels of severity were generated by decreasing the bronchiolar calibers in the right upper lobe of a normal lung (D0). Expiratory flows were recorded in the mid-sagittal plane using a high-speed camera at 1500 fps. In addition to the baseline flow rate (20 L/min) with which the networks were trained and verified, two additional flow rates (15 L/min and 10 L/min) were considered to evaluate the network’s robustness to flow deviations. Distinct flow patterns and vortex dynamics were observed among the three disease states (D0, D1, D2) and across the three flow rates. The AlexNet-LSTM network proved to be robust, maintaining perfect performance in the three-class classification when the flow deviated from the recommendation by 25%, and still performed reasonably (72.8% accuracy) despite a 50% flow deviation. The GoogleNet-LSTM network also showed satisfactory performance (91.5% accuracy) at a 25% flow deviation but exhibited low performance (57.7% accuracy) when the deviation was 50%. Considering the sequential learning effects in this classification task, video classifications only slightly outperformed those using still images (i.e., 3–6%). The occlusion sensitivity analyses showed distinct heat maps specific to the disease state.

Image preprocessing‐based ensemble deep learning classification of diabetic retinopathy

AbstractDiabetic retinopathy (DR) can cause irreversible eye damage, even blindness. The prognosis improves with early diagnosis. According to the International Classification of Diabetic Retinopathy Severity Scale (ICDRSS), DR has five stages. Modern, cost‐effective techniques for automatic DR screening and staging of fundus images are based on deep learning (DL). To obtain higher classification accuracy, the combination of several diverse individual DL models into one ensemble could be used. A new approach to model diversity in an ensemble is proposed by manipulating the training input data involving original and four variants of preprocessed image datasets. There are publicly available datasets with labels for all five stages, but some contain poor‐quality images. In contrast, this algorithm was trained on images from a six‐class DDR dataset, including the class of poor‐quality ungradable images, to enhance the classification performance. The solution was evaluated on the APTOS dataset, containing only ICDRSS classes. Classification results of the ensemble model were presented on two different ensemble convolutional neural network (CNN) models, based on Xception and EfficientNetB4 architectures using two fusion approaches. Our proposed ensemble models outperformed all other single deep learning architectures regarding overall accuracy and Cohen's Kappa, with the best results using the EfficientNetB4 architecture.

Leveraging discriminative features for automatic sleep stage classification based on raw single-channel EEG

Sleep staging is the basis for assessing sleep quality. In the process of scoring each sleep stage, some automatic sleep staging models often fail to effectively capture the more accurate long-range correlation coupling between the input sleep EEG signals and the output sleep stage, which leads to the extracted features cannot effectively distinguish the different sleep stages. We propose an automatic end-to-end sleep stage classification method based on the original single-channel sleep EEG signal to perform feature learning on the source domain information of critical parts of the sleep EEG signals and solve the long-term time series problem. The method uses a convolutional neural network (CNN) to extract the time–frequency domain features of signals. It introduces a squeeze-and-excitation block (SE-Block) on CNN to enhance the feature representation ability of CNN. At the same time, a bidirectional recurrent unit (Bi-GRU) is used to learn the conversion rules of sleep stages, and an attention mechanism is added to the decoding part of Bi-GRU to enhance the long-term memory capacity of Bi-GRU and highlight the influence of essential features. According to the particularity of sleep signals, this method combines multiple models and techniques and creatively blends them to improve the performance of automatic staging. To validate the accuracy and stability of the model, the Fpz-Cz channel and the Pz-Oz channel EEG signals in the Sleep-EDF sleep dataset are used for 10-fold cross-validation. The classification accuracy was 88.48% and 87.56%, respectively. The results show that under the same model architecture and dataset, our model has a more vital ability to extract essential features, better representation ability, more stable performance, and a relatively simple model structure.

A system based on machine learning for improving sleep

Closed-loop auditory stimulation is one of the well-known and emerging sensory stimulation techniques, which achieves the purpose of sleep regulation by driving the EEG slow oscillation (SO, <1 Hz) through auditory stimulation. The main challenge is to accurately identify the stimulation timing and provide feedback in real-time, which has high requirements on the response time and recognition accuracy of the closed-loop auditory stimulation system. To reduce the impact of systematic errors on the regulation results, most traditional closed-loop auditory stimulation systems try to identify a single feature to determine the timing of stimulus delivery and reduce the system feedback delay by simplifying the calculation. Unlike existing closed-loop regulation systems that identify specific brain features, this paper proposes a closed-loop auditory stimulation sleep regulation system deploying machine learning. The process is: through online sleep real-time automatic staging, tracking the sleep stage to provide feedback quickly, and continuously offering external auditory stimulation at a specific SO phase. This paper uses this system to conduct sleep auditory stimulation regulation experiments on ten subjects. The experimental results show that the sleep closed-loop regulation system proposed in this paper can achieve consistency (effective for almost all subjects in the experiment) and immediate (taking effect immediately after stimulation) modulation effects on SOs. More importantly, the proposed method is superior to existing advanced methods. Therefore, the system designed in this paper has great potential to be more reliable and flexible in sleep regulation.

The effect of maxillary molar distalization with clear aligner: a 4D finite-element study with staging simulation

IntroductionLong-term simulation of tooth movement is crucial for clear aligner (CA) treatment. This study aimed to investigate the effect of maxillary molar distalization with CA via an automatic staging simulation.MethodA finite-element method (FEM) model of maxillary dentition, periodontal ligaments, attachments, and corresponding CA was established, and a prescribed 2-mm distalization with 0.1 mm each step of the second molar was simulated. The long-term tooth movement under orthodontic force was simulated with an iterative computation method. The morphologic changes of CA during staging were simulated with the thermal expansion method.ResultsTwenty steps of molar distalization were simulated. Significant distal tilting of the second molar was revealed, along with the proclination of anterior teeth, which caused the ‘reversed bow effect’. For the second molar, 4.63°distal tilting at the 20th step was revealed. The intrusion of the incisors and the second molar were 0.43 mm, 0.39 mm, and 0.45 mm, respectively, at step 20. All the anterior teeth showed a proclination of approximately 1.41°–2.01° at the 20th step. The expression rate of the designed distalization of the second molar was relatively low (approximately 68%) compared to the high efficacy of interdental space opening between molars with CA (approximately 89%).ConclusionA novel method of simulating long-term molar distalization with CA with FEM was developed. The FEM results suggested distal tilting of the second molar and the proclination of anterior teeth during the molar distalization.

Application of artificial intelligence for automatic cataract staging based on anterior segment images: comparing automatic segmentation approaches to manual segmentation.

Cataract is one of the leading causes of blindness worldwide, accounting for >50% of cases of blindness in low- and middle-income countries. In this study, two artificial intelligence (AI) diagnosis platforms are proposed for cortical cataract staging to achieve a precise diagnosis. A total of 647 high quality anterior segment images, which included the four stages of cataracts, were collected into the dataset. They were divided randomly into a training set and a test set using a stratified random-allocation technique at a ratio of 8:2. Then, after automatic or manual segmentation of the lens area of the cataract, the deep transform-learning (DTL) features extraction, PCA dimensionality reduction, multi-features fusion, fusion features selection, and classification models establishment, the automatic and manual segmentation DTL platforms were developed. Finally, the accuracy, confusion matrix, and area under the receiver operating characteristic (ROC) curve (AUC) were used to evaluate the performance of the two platforms. In the automatic segmentation DTL platform, the accuracy of the model in the training and test sets was 94.59 and 84.50%, respectively. In the manual segmentation DTL platform, the accuracy of the model in the training and test sets was 97.48 and 90.00%, respectively. In the test set, the micro and macro average AUCs of the two platforms reached >95% and the AUC for each classification was >90%. The results of a confusion matrix showed that all stages, except for mature, had a high recognition rate. Two AI diagnosis platforms were proposed for cortical cataract staging. The resulting automatic segmentation platform can stage cataracts more quickly, whereas the resulting manual segmentation platform can stage cataracts more accurately.

Performing Automatic Identification and Staging of Urothelial Carcinoma in Bladder Cancer Patients Using a Hybrid Deep-Machine Learning Approach

Accurate clinical staging of bladder cancer aids in optimizing the process of clinical decision-making, thereby tailoring the effective treatment and management of patients. While several radiomics approaches have been developed to facilitate the process of clinical diagnosis and staging of bladder cancer using grayscale computed tomography (CT) scans, the performances of these models have been low, with little validation and no clear consensus on specific imaging signatures. We propose a hybrid framework comprising pre-trained deep neural networks for feature extraction, in combination with statistical machine learning techniques for classification, which is capable of performing the following classification tasks: (1) bladder cancer tissue vs. normal tissue, (2) muscle-invasive bladder cancer (MIBC) vs. non-muscle-invasive bladder cancer (NMIBC), and (3) post-treatment changes (PTC) vs. MIBC.

Study on the method of polysomnography sleep stage staging based on attention mechanism and bidirectional gate recurrent unit

Polysomnography (PSG) monitoring is an important method for clinical diagnosis of diseases such as insomnia, apnea and so on. In order to solve the problem of time-consuming and energy-consuming sleep stage staging of sleep disorder patients using manual frame-by-frame visual judgment PSG, this study proposed a deep learning algorithm model combining convolutional neural networks (CNN) and bidirectional gate recurrent neural networks (Bi GRU). A dynamic sparse self-attention mechanism was designed to solve the problem that gated recurrent neural networks (GRU) is difficult to obtain accurate vector representation of long-distance information. This study collected 143 overnight PSG data of patients from Shanghai Mental Health Center with sleep disorders, which were combined with 153 overnight PSG data of patients from the open-source dataset, and selected 9 electrophysiological channel signals including 6 electroencephalogram (EEG) signal channels, 2 electrooculogram (EOG) signal channels and a single mandibular electromyogram (EMG) signal channel. These data were used for model training, testing and evaluation. After cross validation, the accuracy was (84.0±2.0)%, and Cohen's kappa value was 0.77±0.50. It showed better performance than the Cohen's kappa value of physician score of 0.75±0.11. The experimental results show that the algorithm model in this paper has a high staging effect in different populations and is widely applicable. It is of great significance to assist clinicians in rapid and large-scale PSG sleep automatic staging.

Benchmarking performance of an automatic polysomnography scoring system in a population with suspected sleep disorders.

The current gold standard for measuring sleep disorders is polysomnography (PSG), which is manually scored by a sleep technologist. Scoring a PSG is time-consuming and tedious, with substantial inter-rater variability. A deep-learning-based sleep analysis software module can perform autoscoring of PSG. The primary objective of the study is to validate the accuracy and reliability of the autoscoring software. The secondary objective is to measure workflow improvements in terms of time and cost via a time motion study. The performance of an automatic PSG scoring software was benchmarked against the performance of two independent sleep technologists on PSG data collected from patients with suspected sleep disorders. The technologists at the hospital clinic and a third-party scoring company scored the PSG records independently. The scores were then compared between the technologists and the automatic scoring system. An observational study was also performed where the time taken for sleep technologists at the hospital clinic to manually score PSGs was tracked, along with the time taken by the automatic scoring software to assess for potential time savings. Pearson's correlation between the manually scored apnea-hypopnea index (AHI) and the automatically scored AHI was 0.962, demonstrating a near-perfect agreement. The autoscoring system demonstrated similar results in sleep staging. The agreement between automatic staging and manual scoring was higher in terms of accuracy and Cohen's kappa than the agreement between experts. The autoscoring system took an average of 42.7 s to score each record compared with 4,243 s for manual scoring. Following a manual review of the auto scores, an average time savings of 38.6 min per PSG was observed, amounting to 0.25 full-time equivalent (FTE) savings per year. The findings indicate a potential for a reduction in the burden of manual scoring of PSGs by sleep technologists and may be of operational significance for sleep laboratories in the healthcare setting.

A Developed LSTM-Ladder-Network-Based Model for Sleep Stage Classification.

Sleep staging is crucial for diagnosing sleep-related disorders. The heavy and time-consuming task of manual staging can be released by automatic techniques. However, the automatic staging model would have a relatively poor performance when working on unseen new data due to individual differences. In this research, a developed LSTM-Ladder-Network (LLN) model is proposed for automatic sleep stage classification. Several features are extracted for each epoch and combined with the following epochs to form a cross-epoch vector. The long short-term memory (LSTM) network is added into the basic ladder network (LN) to learn the sequential information of adjacent epochs. The developed model is implemented based on a transductive learning scheme to avoid the issue of accuracy loss caused by individual differences. In this process, the labeled data pre-trains the encoder, and the unlabeled data re- fine the model parameters by minimizing the reconstruction loss. The proposed model is evaluated on the data from public database and hospital. Comparison experiments were conducted where the developed LLN model achieved rather satisfied performance while dealing with the unseen new data. The obtained results demonstrate the effectiveness of the proposed approach in addressing individual differences. This can improve the quality of automatic sleep staging when assessed on different individuals and has strong application potential as a computer aided approach for sleep staging.

EEG-Based Sleep Stage Classification via Neural Architecture Search.

With the improvement of quality of life, people are more and more concerned about the quality of sleep. The electroencephalogram (EEG)-based sleep stage classification is a good guide for sleep quality and sleep disorders. At this stage, most automatic staging neural networks are designed by human experts, and this process is time-consuming and laborious. In this paper, we propose a novel neural architecture search (NAS) framework based on bilevel optimization approximation for EEG-based sleep stage classification. The proposed NAS architecture mainly performs the architectural search through a bilevel optimization approximation, and the model is optimized by search space approximation and search space regularization with parameters shared among cells. Finally, we evaluated the performance of the model searched by NAS on the Sleep-EDF-20, Sleep-EDF-78 and SHHS datasets with an average accuracy of 82.7%, 80.0% and 81.9%, respectively. The experimental results show that the proposed NAS algorithm provides some reference for the subsequent automatic design of networks for sleep classification.

Automatic Sleep Staging Based on Single-Channel EEG Signal Using Null Space Pursuit Decomposition Algorithm

Sleep quality is related to people’s physical and mental health, so an accurate assessment of sleep quality is key to recognizing sleep disorders and taking effective interventions. To address the shortcomings of traditional manual and automatic staging methods, such as being time-consuming and having low classification accuracy, an automatic sleep staging method based on the null space pursuit (NSP) decomposition algorithm of single-channel electroencephalographic (EEG) signals is proposed, which provides a new way for EEG signal decomposition and automatic identification of sleep stages. First, the single-channel EEG signal data from the Sleep-EDF database, DREAMS Subject database, and Sleep Heart Health Study database (SHHS), available on PhysioNet, were preprocessed, respectively. Second, the preprocessed single-channel EEG signals were decomposed by the NSP algorithm. Third, we extracted nine features in the time domain of the nonlinear dynamics and statistics from the original EEG signal and the six simple signals that were decomposed. Finally, the extreme gradient boosting (XGBOOST) algorithm was used to construct a classification model to classify and identify the 63 extracted EEG signal features for automatic sleep staging. The experimental results showed that, on the Sleep-EDF database, the accuracy of four and five categories were 93.59% and 92.89%, respectively; on the DREAMS Subject database, the accuracy rates of four and five categories were 91.32% and 90.01%, respectively; on the SHHS database, the accuracy rates of four and five categories were 90.25% and 88.37%, respectively. The experimental results show that the automatic sleep staging model proposed in this work has high classification accuracy and efficiency, as well as strong applicability and robustness.

A machine learning method for improving liver cancer staging

Liver cancer is a common malignant tumor, and its clinical stage is closely related to the clinical treatment and prognosis of patients. Currently, the BCLC staging system revised by the BCLC group of University of Barcelona is the globally recognized staging system for liver cancer. However, with the deepening of related research, the current staging system can no longer fully meet the clinical needs. In this work, we propose a novel machine learning method for constructing an automatic hepatocellular carcinoma staging model that incorporates far more clinical variables than any existing staging system. Our model is based on random survival forests, which generates a unique hazard function for each patient. B-splines are used to embed hazard functions into vectors in low-dimensional space and hierarchical clustering method groups similar patients to form staging cohorts. The resulting staging system significantly outperforms the BCLC system in terms of distinctiveness between patients in different stages.

CASPHAr: Cache-Managed Accelerator Staging and Pipelining in Heterogeneous System Architectures

Integrating accelerators onto the same chip with CPUs sharing the last level cache (LLC) is beneficial when CPUs and accelerators frequently exchange data. However, if shared data exceeds LLC capacity, expensive spills to and refetches from DRAM will be incurred, limiting the benefits of such integrated architectures. While this can be avoided through careful software optimizations, such as fine-grain data tiling and accelerator synchronization, this involves significant software changes and programmer effort. In this article, we introduce CASPHAr, an LLC architecture that performs automatic, software- and hardware-transparent fine-grain data staging and synchronization between CPUs and accelerators in hardware. CASPHAr tracks and synchronizes producer and consumer accesses at cache line granularity. As soon as some fraction of shared data is produced and becomes ready in the LLC, the data will be delivered for processing in the waiting consumer. Furthermore, CASPHAr extends existing replacement policies to leverage synchronization information for making better eviction decisions. All combined, CASPHAr reduces data spills due to unnecessarily long lifetimes of shared data in the cache. In addition, fine-grain staging and synchronization inherently achieves system-level pipelining of interdependent kernels that can outperform software optimizations. Results show that CASPHAr can boost performance by up to 23% and achieve energy savings of up to 22% over baseline accelerations.

LiSNet: An artificial intelligence -based tool for liver imaging staging of hepatocellular carcinoma aggressiveness.

Presurgical assessment of hepatocellular carcinoma (HCC) aggressiveness can benefit patients' treatment options and prognosis. To develop an artificial intelligence (AI) tool, namely, LiSNet, in the task of scoring and interpreting HCC aggressiveness with computed tomography (CT) imaging. A total of 358 patients with HCC undergoing curative liver resection were retrospectively included. Three subspecialists were recruited to pixel-wise annotate and grade tumor aggressiveness based on CT imaging. LiSNet was trained and validated in 193 and 61 patients with a deep neural network to emulate the diagnostic acumen of subspecialists for staging HCC. The test set comprised 104 independent patients. We subsequently compared LiSNet with an experience-based binary diagnosis scheme and human-AI partnership that combined binary diagnosis and LiSNet for assessing tumor aggressiveness. We also assessed the efficiency of LiSNet for predicting survival outcomes. At the pixel-wise level, the agreement rate of LiSNet with subspecialists was 0.658 (95% confidence interval [CI]: 0.490-0.779), 0.595 (95% CI: 0.406-0.734), and 0.369 (95% CI: 0.134-0.566), for scoring HCC aggressiveness grades I, II, and III, respectively. Additionally, LiSNet was comparable to subspecialists for predicting histopathological microvascular invasion (area under the curve: LiSNet: 0.668 [95% CI: 0.559-0.776] versus subspecialists: 0.699 [95% CI: 0.591-0.806], p>0.05). In a human-AI partnered diagnosis, combining LiSNet and experience-based binary diagnosis can achieve the best predictive ability for microvascular invasion (area under the curve: 0.705 [95% CI: 0.589-0.820]). Furthermore, LiSNet was able to indicate overall survival after surgery. The designed LiSNet tool warrants evaluation as an alternative tool for radiologists to conduct automatic staging of HCC aggressiveness at the pixel-wise level with CT imaging. Its prognostic value might benefit patients' treatment options and survival prediction.

A deep learning-based automatic staging method for early endometrial cancer on MRI images.

Early treatment increases the 5-year survival rate of patients with endometrial cancer (EC). Deep learning (DL) as a new computer-aided diagnosis method has been widely used in medical image processing which can reduce the misdiagnosis by radiologists. An automatic staging method based on DL for the early diagnosis of EC will benefit both radiologists and patients. To develop an effective and automatic prediction model for early EC diagnosis on magnetic resonance imaging (MRI) images, we retrospectively enrolled 117 patients (73 of stage IA, 44 of stage IB) with a pathological diagnosis of early EC confirmed by postoperative biopsy at our institution from 1 January 2018, to 31 December 2020. Axial T2-weighted image (T2WI), axial diffusion-weighted image (DWI) and sagittal T2WI images from 117 patients have been classified into stage IA and stage IB according to the patient’s pathological diagnosis. Firstly, a semantic segmentation model based on the U-net network is trained to segment the uterine region and the tumor region on the MRI images. Then, the area ratio of the tumor region to the uterine region (TUR) in the segmentation map is calculated. Finally, the receiver operating characteristic curves (ROCs) are plotted by the TUR and the results of the patient’s pathological diagnosis in the test set to find the optimal staging thresholds for stage IA and stage IB. In the test sets, the trained semantic segmentation model yields the average Dice similarity coefficients of uterus and tumor on axial T2WI, axial DWI, and sagittal T2WI were 0.958 and 0.917, 0.956 and 0.941, 0.972 and 0.910 respectively. With pathological diagnostic results as the gold standard, the classification model on axial T2WI, axial DWI, and sagittal T2WI yielded an area under the curve (AUC) of 0.86, 0.85 and 0.94, respectively. In this study, an automatic DL-based segmentation model combining the ROC analysis of TUR on MRI images presents an effective early EC staging method.

Intelligent automatic sleep staging model based on CNN and LSTM.

Since electroencephalogram (EEG) is a significant basis to treat and diagnose somnipathy, sleep electroencephalogram automatic staging methods play important role in the treatment and diagnosis of sleep disorders. Due to the characteristics of weak signals, EEG needs accurate and efficient algorithms to extract feature information before applying it in the sleep stages. Conventional feature extraction methods have low efficiency and are difficult to meet the time validity of fast staging. In addition, it can easily lead to the omission of key features owing to insufficient a priori knowledge. Deep learning networks, such as convolutional neural networks (CNNs), have powerful processing capabilities in data analysis and data mining. In this study, a deep learning network is introduced into the study of the sleep stage. In this study, the feature fusion method is presented, and long-term and short-term memory (LSTM) is selected as the classification network to improve the accuracy of sleep stage recognition. First, based on EEG and deep learning network, an automatic sleep phase method based on a multi-channel EGG is proposed. Second, CNN-LSTM is used to monitor EEG and EOG samples during sleep. In addition, without any signal preprocessing or feature extraction, data expansion (DA) can be realized for unbalanced data, and special data and non-general data can be deleted. Finally, the MIT-BIH dataset is used to train and evaluate the proposed model. The experimental results show that the EEG-based sleep phase method proposed in this paper provides an effective method for the diagnosis and treatment of sleep disorders, and hence has a practical application value.

Combining Deep Learning and Radiomics for Automated, Objective, Comprehensive Bone Marrow Characterization From Whole-Body MRI: A Multicentric Feasibility Study.

Disseminated bone marrow (BM) involvement is frequent in multiple myeloma (MM). Whole-body magnetic resonance imaging (wb-MRI) enables to evaluate the whole BM. Reading of such whole-body scans is time-consuming, and yet radiologists can transfer only a small fraction of the information of the imaging data set to the report. This limits the influence that imaging can have on clinical decision-making and in research toward precision oncology. The objective of this feasibility study was to implement a concept for automatic, comprehensive characterization of the BM from wb-MRI, by automatic BM segmentation and subsequent radiomics analysis of 30 different BM spaces (BMS). This retrospective multicentric pilot study used a total of 106 wb-MRI from 102 patients with (smoldering) MM from 8 centers. Fifty wb-MRI from center 1 were used for training of segmentation algorithms (nnU-Nets) and radiomics algorithms. Fifty-six wb-MRI from 8 centers, acquired with a variety of different MRI scanners and protocols, were used for independent testing. Manual segmentations of 2700 BMS from 90 wb-MRI were performed for training and testing of the segmentation algorithms. For each BMS, 296 radiomics features were calculated individually. Dice score was used to assess similarity between automatic segmentations and manual reference segmentations. The "multilabel nnU-Net" segmentation algorithm, which performs segmentation of 30 BMS and labels them individually, reached mean dice scores of 0.88 ± 0.06/0.87 ± 0.06/0.83 ± 0.11 in independent test sets from center 1/center 2/center 3-8 (interrater variability between radiologists, 0.88 ± 0.01). The subset from the multicenter, multivendor test set (center 3-8) that was of high imaging quality was segmented with high precision (mean dice score, 0.87), comparable to the internal test data from center 1. The radiomic BM phenotype consisting of 8880 descriptive parameters per patient, which result from calculation of 296 radiomics features for each of the 30 BMS, was calculated for all patients. Exemplary cases demonstrated connections between typical BM patterns in MM and radiomic signatures of the respective BMS. In plausibility tests, predicted size and weight based on radiomics models of the radiomic BM phenotype significantly correlated with patients' actual size and weight ( P = 0.002 and P = 0.003, respectively). This pilot study demonstrates the feasibility of automatic, objective, comprehensive BM characterization from wb-MRI in multicentric data sets. This concept allows the extraction of high-dimensional phenotypes to capture the complexity of disseminated BM disorders from imaging. Further studies need to assess the clinical potential of this method for automatic staging, therapy response assessment, or prediction of biopsy results.