Recent advancements in statistical and reinforcement learning methods have contributed to superior patient care strategies. However, these methods face substantial challenges in high-stakes contexts, including missing data, stochasticity, and the need for interpretability and patient safety. Our work operationalizes a safe and interpretable approach for optimizing treatment regimes by matching patients with similar medical and pharmacological profiles. This allows us to construct optimal policies via interpolation. Our comprehensive simulation study demonstrates our method’s effectiveness in complex scenarios. We use this approach to study seizure treatment in critically ill patients, advocating for personalized strategies based on medical history and pharmacological features. Our findings recommend reducing medication doses for mild, brief seizure episodes and adopting aggressive treatment strategies for severe cases, leading to improved outcomes.

Biological signals, such as electroencephalograms (EEG), play a crucial role in numerous clinical applications, exhibiting diverse data formats and quality profiles. Current deep learning models for biosignals (based on CNN, RNN, and Transformers) are typically specialized for specific datasets and clinical settings, limiting their broader applicability. This paper explores the development of a flexible biosignal encoder architecture that can enable pre-training on multiple datasets and fine-tuned on downstream biosignal tasks with different formats. To overcome the unique challenges associated with biosignals of various formats, such as mismatched channels, variable sample lengths, and prevalent missing val- ues, we propose Biosignal Transformer (BIOT). The proposed BIOT model can enable cross-data learning with mismatched channels, variable lengths, and missing values by tokenizing different biosignals into unified "sentences" structure. Specifically, we tokenize each channel separately into fixed-length segments containing local signal features and then rearrange the segments to form a long "sentence". Channel embeddings and relative position embeddings are added to each segment (viewed as "token") to preserve spatio-temporal features. The BIOT model is versatile and applicable to various biosignal learning settings across different datasets, including joint pre-training for larger models. Comprehensive evaluations on EEG, electrocardiogram (ECG), and human activity sensory signals demonstrate that BIOT outperforms robust baselines in common settings and facilitates learning across multiple datasets with different formats. Using CHB-MIT seizure detection task as an example, our vanilla BIOT model shows 3% improvement over baselines in balanced accuracy, and the pre-trained BIOT models (optimized from other data sources) can further bring up to 4% improvements. Our repository is public at https://github.com/ycq091044/BIOT.

Neurologic disability level at hospital discharge is an important outcome in many clinical research studies. Outside of clinical trials, neurologic outcomes must typically be extracted by labor intensive manual review of clinical notes in the electronic health record (EHR). To overcome this challenge, we set out to develop a natural language processing (NLP) approach that automatically reads clinical notes to determine neurologic outcomes, to make it possible to conduct larger scale neurologic outcomes studies. We obtained 7314 notes from 3632 patients hospitalized at two large Boston hospitals between January 2012 and June 2020, including discharge summaries (3485), occupational therapy (1472) and physical therapy (2357) notes. Fourteen clinical experts reviewed notes to assign scores on the Glasgow Outcome Scale (GOS) with 4 classes, namely 'good recovery', 'moderate disability', 'severe disability', and 'death' and on the Modified Rankin Scale (mRS), with 7 classes, namely 'no symptoms', 'no significant disability', 'slight disability', 'moderate disability', 'moderately severe disability', 'severe disability', and 'death'. For 428 patients' notes, 2 experts scored the cases generating interrater reliability estimates for GOS and mRS. After preprocessing and extracting features from the notes, we trained a multiclass logistic regression model using LASSO regularization and 5-fold cross validation for hyperparameter tuning. The model performed well on the test set, achieving a micro average area under the receiver operating characteristic and F-score of 0.94 (95% CI 0.93-0.95) and 0.77 (0.75-0.80) for GOS, and 0.90 (0.89-0.91) and 0.59 (0.57-0.62) for mRS, respectively. Our work demonstrates that an NLP algorithm can accurately assign neurologic outcomes based on free text clinical notes. This algorithm increases the scale of research on neurological outcomes that is possible with EHR data.

Despite deep learning (DL) success in classification problems, DL classifiers do not provide a sound mechanism to decide when to refrain from predicting. Recent works tried to control the overall prediction risk with classification with rejection options. However, existing works overlook the different significance of different classes. We introduce Set-classifier with class-specific RIsk Bounds (SCRIB) to tackle this problem, assigning multiple labels to each example. Given the output of a black-box model on the validation set, SCRIB constructs a set-classifier that controls the class-specific prediction risks. The key idea is to reject when the set classifier returns more than one label. We validated SCRIB on several medical applications, including sleep staging on electroencephalogram(EEG) data, X-ray COVID image classification, and atrial fibrillation detection based on electrocardiogram (ECG) data.SCRIB obtained desirable class-specific risks, which are 35%-88% closer to the target risks than baseline methods.

Tensor decompositions are powerful tools for dimensionality reduction and feature interpretation of multidimensional data such as signals. Existing tensor decomposition objectives (e.g., Frobenius norm) are designed for fitting raw data under statistical assumptions, which may not align with downstream classification tasks. In practice, raw input tensor can contain irrelevant information while data augmentation techniques may be used to smooth out class-irrelevant noise in samples. This paper addresses the above challenges by proposing augmented tensor decomposition (ATD), which effectively incorporates data augmentations and self-supervised learning (SSL) to boost downstream classification. To address the non-convexity of the new augmented objective, we develop an iterative method that enables the optimization to follow an alternating least squares (ALS) fashion. We evaluate our proposed ATD on multiple datasets. It can achieve 0.8%~2.5% accuracy gain over tensor-based baselines. Also, our ATD model shows comparable or better performance (e.g., up to 15% in accuracy) over self-supervised and autoencoder baselines while using less than 5% of learnable parameters of these baseline models.

Epilepsy diagnosis based on Interictal Epileptiform Discharges (IEDs) in scalp electroencephalograms (EEGs) is laborious and often subjective. Therefore, it is necessary to build an effective IED detector and an automatic method to classify IED-free versus IED EEGs. In this study, we evaluate features that may provide reliable IED detection and EEG classification. Specifically, we investigate the IED detector based on convolutional neural network (ConvNet) with different input features (temporal, spectral, and wavelet features). We explore different ConvNet architectures and types, including 1D (one-dimensional) ConvNet, 2D (two-dimensional) ConvNet, and noise injection at various layers. We evaluate the EEG classification performance on five independent datasets. The 1D ConvNet with preprocessed full-frequency EEG signal and frequency bands (delta, theta, alpha, beta) with Gaussian additive noise at the output layer achieved the best IED detection results with a false detection rate of 0.23/min at 90% sensitivity. The EEG classification system obtained a mean EEG classification Leave-One-Institution-Out (LOIO) cross-validation (CV) balanced accuracy (BAC) of 78.1% (area under the curve (AUC) of 0.839) and Leave-One-Subject-Out (LOSO) CV BAC of 79.5% (AUC of 0.856). Since the proposed classification system only takes a few seconds to analyze a 30-min routine EEG, it may help in reducing the human effort required for epilepsy diagnosis.

The diagnosis of epilepsy often relies on a reading of routine scalp electroencephalograms (EEGs). Since seizures are highly unlikely to be detected in a routine scalp EEG, the primary diagnosis depends heavily on the visual evaluation of Interictal Epileptiform Discharges (IEDs). This process is tedious, expert-centered, and delays the treatment plan. Consequently, the development of an automated, fast, and reliable epileptic EEG diagnostic system is essential. In this study, we propose a system to classify EEG as epileptic or normal based on multiple modalities extracted from the interictal EEG. The ensemble system consists of three components: a Convolutional Neural Network (CNN)-based IED detector, a Template Matching (TM)-based IED detector, and a spectral feature-based classifier. We evaluate the system on datasets from six centers from the USA, Singapore, and India. The system yields a mean Leave-One-Institution-Out (LOIO) cross-validation (CV) area under curve (AUC) of 0.826 (balanced accuracy (BAC) of 76.1%) and Leave-One-Subject-Out (LOSO) CV AUC of 0.812 (BAC of 74.8%). The LOIO results are found to be similar to the interrater agreement (IRA) reported in the literature for epileptic EEG classification. Moreover, as the proposed system can process routine EEGs in a few seconds, it may aid the clinicians in diagnosing epilepsy efficiently.

Generating clinical reports from raw recordings such as X-rays and electroencephalogram (EEG) is an essential and routine task for doctors. However, it is often time-consuming to write accurate and detailed reports. Most existing methods try to generate the whole reports from the raw input with limited success because 1) generated reports often contain errors that need manual review and correction, 2) it does not save time when doctors want to write additional information into the report, and 3) the generated reports are not customized based on individual doctors’ preference. We propose CLinicAl Report Auto-completion (, an interactive method that generates reports in a sentence by sentence fashion based on doctors’ anchor words and partially completed sentences. earches for most relevant sentences from existing reports as the template for the current report. The retrieved sentences are sequentially modified by combining with the input feature representations to create the final report. In our experimental evaluation chieved 0.393 CIDEr and 0.248 BLEU-4 on X-ray reports and 0.482 CIDEr and 0.491 BLEU-4 for EEG reports for sentence-level generation, which is up to 35% improvement over the best baseline. Also via our qualitative evaluation, s shown to produce reports which have a significantly higher level of approval by doctors in a user study (3.74 out of 5 for s 2.52 out of 5 for the baseline).

Visual evaluation of electroencephalogram (EEG) for Interictal Epileptiform Discharges (IEDs) as distinctive biomarkers of epilepsy has various limitations, including time-consuming reviews, steep learning curves, interobserver variability, and the need for specialized experts. The development of an automated IED detector is necessary to provide a faster and reliable diagnosis of epilepsy. In this paper, we propose an automated IED detector based on Convolutional Neural Networks (CNNs). We have evaluated the proposed IED detector on a sizable database of 554 scalp EEG recordings (84 epileptic patients and 461 nonepileptic subjects) recorded at Massachusetts General Hospital (MGH), Boston. The proposed CNN IED detector has achieved superior performance in comparison with conventional methods with a mean cross-validation area under the precision-recall curve (AUPRC) of 0.838[Formula: see text]±[Formula: see text]0.040 and false detection rate of 0.2[Formula: see text]±[Formula: see text]0.11 per minute for a sensitivity of 80%. We demonstrated the proposed system to be noninferior to 30 neurologists on a dataset from the Medical University of South Carolina (MUSC). Further, we clinically validated the system at National University Hospital (NUH), Singapore, with an agreement accuracy of 81.41% with a clinical expert. Moreover, the proposed system can be applied to EEG recordings with any arbitrary number of channels.

In order to reduce the possibility of neural injury from seizures and sidestep the need for a neurologist to spend hours on manually reviewing the EEG recording, it is critical to automatically detect and classify "interictal-ictal continuum" (IIC) patterns from EEG data. However, the existing IIC classification techniques are shown to be not accurate and robust enough for clinical use because of the lack of high quality labels of EEG segments as training data. Obtaining high-quality labeled data is traditionally a manual process by trained clinicians that can be tedious, time-consuming, and error-prone. In this work, we propose Smile, an industrial scale system that provides an end-to-end solution to the IIC pattern classification problem. The core components of Smile include a visualization-based time series labeling module and a deep-learning based active learning module. The labeling module enables the users to explore and label 350 million EEG segments (30TB) at interactive speed. The multiple coordinated views allow the users to examine the EEG signals from both time domain and frequency domain simultaneously. The active learning module first trains a deep neural network that automatically extracts both the local features with respect to each segment itself and the long term dynamics of the EEG signals to classify IIC patterns. Then leveraging the output of the deep learning model, the EEG segments that can best improve the model are selected and prompted to clinicians to label. This process is iterated until the clinicians and the models show high degree of agreement. Our initial experimental results show that our Smile system allows the clinicians to label the EEG segments at will with a response time below 500 ms. The accuracy of the model is progressively improved as more and more high quality labels are acquired over time.

In Intensive Care Unit, the sedation level of patients is usually monitored by periodically assessing the behavioral response to stimuli. However, these clinical assessments are limited due to the disruption with patients' sleep and the noise of observing behaviors instead of the brain activity directly. Here we train a Gated Recurrent Unit using the spectrogram of electroencephalography (EEG) based on 166 mechanically ventilated patients to predict the Richmond Agitation-Sedation Score, scored as ordinal levels of -5, -4, ... up to 0. The model is able to predict 50% accurate with an error not larger than 1 level; and 80% accurate with an error not larger than 2 levels on hold-out testing patients. We show typical spectrograms in each sedation level and interpret the results based on the visualization of the gradient with respect to the spectrogram. Future improvements include utilizing the EEG waveforms since waveform patterns are clinically thought to be associated with sedation levels, as well as training patient-specific models.

The presence of Epileptiform Transients (ET) in the electroencephalogram (EEG) is a key finding in the medical workup of a patient with suspected epilepsy. Automated ET detection can increase the uniformity and speed of ET detection. Current ET detection methods suffer from insufficient precision and high false positive rates. Since ETs occur infrequently in the EEG of most patients, the majority of recordings comprise background EEG waveforms. In this work we establish a method to exclude as much background data as possible from EEG recordings by applying a classifier cascade. The remaining data can then be classified using other ET detection methods. We compare a single Support Vector Machine (SVM) to a cascade of SVMs for detecting ETs. Our results show that the precision and false positive rate improve significantly by incorporating a classifier cascade before ET detection. Our method can help improve the precision and false positive rate of an ET detection system. At a fixed sensitivity, we were able to improve precision by 6.78%; and at a fixed false positive rate, the sensitivity improved by 2.83%.

Objectives: Scoring laboratory polysomnography (PSG) data remains a manual task of visually annotating 3 primary categories: sleep stages, sleep disordered breathing, and limb movements. Attempts to automate this process have been hampered by the complexity of PSG signals and physiological heterogeneity between patients. Deep neural networks, which have recently achieved expert-level performance for other complex medical tasks, are ideally suited to PSG scoring, given sufficient training data. Methods: We used a combination of deep recurrent and convolutional neural networks (RCNN) for supervised learning of clinical labels designating sleep stages, sleep apnea events, and limb movements. The data for testing and training were derived from 10 000 clinical PSGs and 5804 research PSGs. Results: When trained on the clinical dataset, the RCNN reproduces PSG diagnostic scoring for sleep staging, sleep apnea, and limb movements with accuracies of 87.6%, 88.2% and 84.7% on held-out test data, a level of performance comparable to human experts. The RCNN model performs equally well when tested on the independent research PSG database. Only small reductions in accuracy were noted when training on limited channels to mimic at-home monitoring devices: frontal leads only for sleep staging, and thoracic belt signals only for the apnea-hypopnea index. Conclusions: By creating accurate deep learning models for sleep scoring, our work opens the path toward broader and more timely access to sleep diagnostics. Accurate scoring automation can improve the utility and efficiency of in-lab and at-home approaches to sleep diagnostics, potentially extending the reach of sleep expertise beyond specialty clinics.

Clinical researchers, historians, educators and field researchers alike still regularly capture data on paper spreadsheets. In the case of health care and education, data will often contain sensitive personal information, further complicating the process of transcribing paper-based archives into digital form. In this work, we describe a tool that utilizes machine learning and crowd intelligence to automatically transcribe images of paper-based spreadsheets into electronic form while protecting sensitive personal information. Our solution consists of four high-level stages: (1) the extraction of cell-level images from the spreadsheet grid, (2) machine recognition of digits within the cells, (3) human transcription of cell contents that the machine was uncertain of and (4) feedback of human transcription results to the machine to improve future classification performance. We test the algorithm on a novel data-set of 135 heterogeneous clinical flow-sheet images collected from the Massachusetts General Hospital (MGH), 2 hand-drawn spreadsheets, one chalk-board drawing, and one printed table. we demonstrate that our algorithm provides a generalized solution for spreadsheet transcription that maintains privacy, is up to 10 times faster and twice as cost effective than existing alternatives. Our work is valuable both as a tool and as a starting point for the development of better algorithms.

Automated annotation of electroencephalograms (EEG) of epileptic patients is important in diagnosis and management of epilepsy. Epilepsy is often associated with the presence of epileptiform transients (ET) in the EEG. To develop an efficient ET detector, a vast amount of data is required to train and evaluate the performance of the detector. Interictal EEG data contains mostly background waveforms, since ETs only occur occasionally in most patients. In order to detect ETs in an automated fashion, it is meaningful to first try to eliminate most background waveforms by means of simple, fast classifiers. The remaining waveforms can in a following step be processed by more sophisticated and computationally demanding classification algorithms, such as deep learning systems. In this study, we design a cascade of simple thresholding steps to reject most background waveforms in interictal EEG, while maintaining most ETs. Several simple and quick-to-compute EEG features are chosen. By thresholding these features in consecutive steps, background waveforms are rejected sequentially. In our numerical experiments, a cascade of 10 steps is able to reject 98.65% of all background segments in the dataset, while preserving 90.6% of the ETs.

The EEG of epileptic patients often contains sharp waveforms called "spikes", occurring between seizures. Detecting such spikes is crucial for diagnosing epilepsy. In this paper, we develop a convolutional neural network (CNN) for detecting spikes in EEG of epileptic patients in an automated fashion. The CNN has a convolutional architecture with filters of various sizes applied to the input layer, leaky ReLUs as activation functions, and a sigmoid output layer. Balanced mini-batches were applied to handle the imbalance in the data set. Leave-one-patient-out cross-validation was carried out to test the CNN and benchmark models on EEG data of five epilepsy patients. We achieved 0.947 AUC for the CNN, while the best performing benchmark model, Support Vector Machines with Gaussian kernel, achieved an AUC of 0.912.

Epilepsy is often associated with the presence of spikes in electroencephalograms (EEGs). The spike waveforms vary vastly among epilepsy patients, and also for the same patient across time. In order to develop semi-automated and automated methods for detecting spikes, it is crucial to obtain a better understanding of the various spike shapes. In this paper, we develop several approaches to extract exemplars of spikes. We generate spike exemplars by applying clustering algorithms to a database of spikes from 12 patients. As similarity measures for clustering, we consider the Euclidean distance and Dynamic Time Warping (DTW). We assess two clustering algorithms, namely, K-means clustering and affinity propagation. The clustering methods are compared based on the mean squared error, and the similarity measures are assessed based on the number of generated spike clusters. Affinity propagation with DTW is shown to be the best combination for clustering epileptic spikes, since it generates fewer spike templates and does not require to pre-specify the number of spike templates.

Extreme learning machine (ELM) was initially proposed for single-hidden-layer feedforward neural networks (SLFNs). In the hidden layer (feature mapping), nodes are randomly generated independently of training data. Furthermore, a unified ELM was proposed, providing a single framework to simplify and unify different learning methods, such as SLFNs, least square support vector machines, proximal support vector machines, and so on. However, the solution of unified ELM is dense, and thus, usually plenty of storage space and testing time are required for large-scale applications. In this paper, a sparse ELM is proposed as an alternative solution for classification, reducing storage space and testing time. In addition, unified ELM obtains the solution by matrix inversion, whose computational complexity is between quadratic and cubic with respect to the training size. It still requires plenty of training time for large-scale problems, even though it is much faster than many other traditional methods. In this paper, an efficient training algorithm is specifically developed for sparse ELM. The quadratic programming problem involved in sparse ELM is divided into a series of smallest possible sub-problems, each of which are solved analytically. Compared with SVM, sparse ELM obtains better generalization performance with much faster training speed. Compared with unified ELM, sparse ELM achieves similar generalization performance for binary classification applications, and when dealing with large-scale binary classification problems, sparse ELM realizes even faster training speed than unified ELM.

Most time series data mining algorithms use similarity search as a core subroutine, and thus the time taken for similarity search is the bottleneck for virtually all time series data mining algorithms. The difficulty of scaling search to large datasets largely explains why most academic work on time series data mining has plateaued at considering a few millions of time series objects, while much of industry and science sits on billions of time series objects waiting to be explored. In this work we show that by using a combination of four novel ideas we can search and mine truly massive time series for the first time. We demonstrate the following extremely unintuitive fact; in large datasets we can exactly search under DTW much more quickly than the current state-of-the-art Euclidean distance search algorithms. We demonstrate our work on the largest set of time series experiments ever attempted. In particular, the largest dataset we consider is larger than the combined size of all of the time series datasets considered in all data mining papers ever published. We show that our ideas allow us to solve higher-level time series data mining problem such as motif discovery and clustering at scales that would otherwise be untenable. In addition to mining massive datasets, we will show that our ideas also have implications for real-time monitoring of data streams, allowing us to handle much faster arrival rates and/or use cheaper and lower powered devices than are currently possible.

Time series motifs are sets of very similar subsequences of a long time series. They are of interest in their own right, and are also used as inputs in several higher-level data mining algorithms including classification, clustering, rule-discovery and summarization. In spite of extensive research in recent years, finding time series motifs exactly in massive databases is an open problem. Previous efforts either found approximate motifs or considered relatively small datasets residing in main memory . In this work, we leverage off previous work on pivot-based indexing to introduce a disk-aware algorithm to find time series motifs exactly in multi-gigabyte databases which contain on the order of tens of millions of time series. We have evaluated our algorithm on datasets from diverse areas including medicine, anthropology, computer networking and image processing and show that we can find interesting and meaningful motifs in datasets that are many orders of magnitude larger than anything considered before.

Time series motifs are pairs of individual time series, or subsequences of a longer time series, which are very similar to each other. As with their discrete analogues in computational biology, this similarity hints at structure which has been conserved for some reason and may therefore be of interest. Since the formalism of time series motifs in 2002, dozens of researchers have used them for diverse applications in many different domains. Because the obvious algorithm for computing motifs is quadratic in the number of items, more than a dozen approximate algorithms to discover motifs have been proposed in the literature. In this work, for the first time, we show a tractable exact algorithm to find time series motifs. As we shall show through extensive experiments, our algorithm is up to three orders of magnitude faster than brute-force search in large datasets. We further show that our algorithm is fast enough to be used as a subroutine in higher level data mining algorithms for anytime classification, near-duplicate detection and summarization, and we consider detailed case studies in domains as diverse as electroencephalograph interpretation and entomological telemetry data mining.

We present a simple geometrical interpretation for the solution to the multiple hypothesis testing problem in the asymptotic limit. Under this interpretation, the optimal decision rule is a nearest neighbor classifier on the probability simplex.

Biological and machine pattern recognition systems face a common challenge: Given sensory data about an unknown pattern, classify the pattern by searching for the best match within a library of representations stored in memory. In many cases, the number of patterns to be discriminated and the richness of the raw data force recognition systems to internally represent memory and sensory information in a compressed format. However, these representations must preserve enough information to accommodate the variability and complexity of the environment, otherwise recognition will be unreliable. Thus, there is an intrinsic tradeoff between the amount of resources devoted to data representation and the complexity of the environment in which a recognition system may reliably operate. In this paper, we describe a mathematical model for pattern recognition systems subject to resource constraints, and show how the aforementioned resource-complexity tradeoff can be characterized in terms of three rates related to the number of bits available for representing memory and sensory data, and the number of patterns populating a given statistical environment. We prove single-letter information-theoretic bounds governing the achievable rates, and investigate in detail two illustrative cases where the pattern data is either binary or Gaussian.

In this paper we examine the achievable rate region for the problem of successive refinement of information for pattern recognition systems. The pattern recognition system has two stages, going from coarse to fine recognition as more resources become available for storing internal representations of the patterns. We present an inner and an outer bound on the true achievable rate region. Using these results we derive conditions under which a pattern recognition system is successively refinable. These conditions are similar to the Markov condition for successive refinement in the rate-distortion problem.

Recently we presented information-theoretic bounds for the achievable rates of pattern recognition systems operating under data compression constraints. In this paper we improve on our previous inner bound, and report progress toward finding formulas for the achievable rate region boundaries in the special cases where the pattern data is either binary or Gaussian

We propose a model for rate-constrained pattern recognition problems, and present single-letter information bounds governing the conditions under which asymptotically error-free recognition is possible. The bounds depend on the statistics of the training and testing data, the number of pattern classes, and the rates of the codes used by the recognition system to internalize the data.