The United States, as well as the global community, experiences health disparities among socially disadvantaged populations. These disparities often manifest in the data utilized for AI model training. Without appropriate de-biasing strategies, models trained to optimize predictive performance may inadvertently capture and perpetuate these inherent biases. The utilization of biased models in clinical decision-making can inflict harm upon patients from disadvantaged groups and exacerbate disparities when these decisions are documented and employed to train subsequent AI models. Unlike conventionalcorrelation-basedmethods, we aim to mitigate the negative impacts of health disparity by answering acausal inferencequestion for fairness:would the clinical decision support system make a different decision if the patient had a different sensitive attribute (e.g., race)?Recognizing the high computational complexity of developing causal models, we propose a flexible and efficient causal-model-free algorithm,CFReg, which provides causal fairness for supervised machine learning models. In addition,CFRegalso develops a novel evaluation metric to quantify fairness within clinical settings. We first validateCFRegusing a healthcare dataset of 48,784 patients focused on care management, then generalize to another four benchmark datasets with racial and ethnic disparity, including law school admission, adult income, criminal recidivism, and violent crime prediction. Experimental results demonstrate thatCFRegoutperforms baseline approaches in both fairness and accuracy, achieving a good trade-off between model fairness and supervised classification performance.

Clinicians often rely on data engineers to retrieve complex patient information from electronic health record (EHR) systems, a process that is both inefficient and time-consuming. We propose EHRAgent, a large language model (LLM) agent empowered with accumulative domain knowledge and robust coding capability. EHRAgent enables autonomous code generation and execution to facilitate clinicians in directly interacting with EHRs using natural language. Specifically, we formulate a multi-tabular reasoning task based on EHRs as a tool-use planning process, efficiently decomposing a complex task into a sequence of manageable actions with external toolsets. We first inject relevant medical information to enable EHRAgent to effectively reason about the given query, identifying and extracting the required records from the appropriate tables. By integrating interactive coding and execution feedback, EHRAgent then effectively learns from error messages and iteratively improves its originally generated code. Experiments on three real-world EHR datasets show that EHRAgent outperforms the strongest baseline by up to 29.6% in success rate, verifying its strong capacity to tackle complex clinical tasks with minimal demonstrations.

Supervised machine learning algorithms depend on accurate labeling of target data to develop models that can derive relationships between input data and the target data. One major hindrance for developing supervised machine learning models capable of predicting the correct target label of unseen data rests on the quality of the data used to train the models, which often depends on having a subject matter expert (SME) create a labeled dataset to train the model on. Given the scarcity of such experts in many fields, the time needed to analyze data for labeling, and subjective differences among experts, ways to reduce the complexity associated with creating meaningful datasets are needed. In this work, we explore the use of two unsupervised topic modeling algorithms, Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorization (NMF) as potential methods for reducing the complexities in the labeling process. Specifically, we obtained COVID patient message data labeled by a SME and compared the overlap in topics designated as COVID versus not by the two algorithms to those of the SME. For each of the topic modeling algorithms, we found a strong degree of overlap in the COVID vs. non-COVID patient message labels with that of the SME, suggesting that the methodology could be used to provide synergies for developing labeled data sets used for clinically meaningful models.

Expert microscopic analysis of cells obtained from frequent heart biopsies is vital for early detection of pediatric heart transplant rejection to prevent heart failure. Detection of this rare condition is prone to low levels of expert agreement due to the difficulty of identifying subtle rejection signs within biopsy samples. The rarity of pediatric heart transplant rejection also means that very few gold-standard images are available for developing machine learning models. To solve this urgent clinical challenge, we developed a deep learning model to automatically quantify rejection risk within digital images of biopsied tissue using an explainable synthetic data augmentation approach. We developed this explainable AI framework to illustrate how our progressive and inspirational generative adversarial network models distinguish between normal tissue images and those containing cellular rejection signs. To quantify biopsy-level rejection risk, we first detect local rejection features using a binary image classifier trained with expert-annotated and synthetic examples. We converted these local predictions into a biopsy-wide rejection score via an interpretable histogram-based approach. Our model significantly improves upon prior works with the same dataset with an area under the receiver operating curve (AUROC) of 98.84% for the local rejection detection task and 95.56% for the biopsy-rejection prediction task. A biopsy-level sensitivity of 83.33% makes our approach suitable for early screening of biopsies to prioritize expert analysis. Our framework provides a solution to rare medical imaging challenges currently limited by small datasets.

Each year there are nearly 57 million deaths worldwide, with over 2.7 million in the United States. Timely, accurate and complete death reporting is critical for public health, especially during the COVID-19 pandemic, as institutions and government agencies rely on death reports to formulate responses to communicable diseases. Unfortunately, determining the causes of death is challenging even for experienced physicians. The novel coronavirus and its variants may further complicate the task, as physicians and experts are still investigating COVID-related complications. To assist physicians in accurately reporting causes of death, an advanced Artificial Intelligence (AI) approach is presented to determine a chronically ordered sequence of conditions that lead to death (named as the causal sequence of death), based on decedent's last hospital discharge record. The key design is to learn the causal relationship among clinical codes and to identify death-related conditions. There exist three challenges: different clinical coding systems, medical domain knowledge constraint, and data interoperability. First, we apply neural machine translation models with various attention mechanisms to generate sequences of causes of death. We use the BLEU (BiLingual Evaluation Understudy) score with three accuracy metrics to evaluate the quality of generated sequences. Second, we incorporate expert-verified medical domain knowledge as constraints when generating the causal sequences of death. Lastly, we develop a Fast Healthcare Interoperability Resources (FHIR) interface that demonstrates the usability of this work in clinical practice. Our results match the state-of-art reporting and can assist physicians and experts in public health crisis such as the COVID-19 pandemic.

Scoliosis is a spinal curvature that most frequently affects adolescents. Posterior spinal fusion surgery is required to correct the deformity in patients with severe scoliosis. Surgeons frequently use radiographic measurements and patient reported outcomes to aid in surgical treatment and monitor patient rehabilitation. Shriners Hospitals for Children is a large healthcare system caring for a significant percentage of pediatric patients with scoliosis. Surgeons from SHC-Greenville and SHC-Lexington have recorded data from more than 1,000 individual scoliosis patients. However, these collected data are usually dispersed across individual healthcare sites, necessitating the development of an integrated clinical data repository for data sharing and management. In this paper, we established a standardized research data repository with FHIR resources to harmonize multi-modal patient data from multiple clinical sites. Additionally, a FHIR-compliant application with a web-based user interface was prototyped to enable clinicians and researchers to access scoliosis patient data within our integrated and standardized research repository. Patient cohort definitions can be used to search these records using the same FHIR application. This standardized data-sharing framework and healthcare information system can be applied to multi-site and multimodality studies for clinical and research purposes, with the ultimate goal of improving the quality of patient care.

Researchers seek help from deep learning methods to alleviate the enormous burden of reading radiological images by clinicians during the COVID-19 pandemic. However, clinicians are often reluctant to trust deep models due to their black-box characteristics. To automatically differentiate COVID-19 and community-acquired pneumonia from healthy lungs in radiographic imaging, we propose an explainable attention-transfer classification model based on the knowledge distillation network structure. The attention transfer direction always goes from the teacher network to the student network. Firstly, the teacher network extracts global features and concentrates on the infection regions to generate attention maps. It uses a deformable attention module to strengthen the response of infection regions and to suppress noise in irrelevant regions with an expanded reception field. Secondly, an image fusion module combines attention knowledge transferred from teacher network to student network with the essential information in original input. While the teacher network focuses on global features, the student branch focuses on irregularly shaped lesion regions to learn discriminative features. Lastly, we conduct extensive experiments on public chest X-ray and CT datasets to demonstrate the explainability of the proposed architecture in diagnosing COVID-19.

Electroencephalogram (EEG) signals can be used for seizure detection, but the seizure patterns found in between patient's EEGs can have significant variations. Specifically, focal spikes in patient-specific channels as well as other patient specific patterns can strongly indicate seizure activity. Manual diagnosis on these markers leads to inconsistent interrater agreement and poor detection accuracy. Previous automation attempts have ignored patient specific approaches but fail to generalize to previously unseen patients. To reduce subjectivity in manual diagnosis, we propose an automatic seizure detection pipeline that includes quality control, preprocessing, and meta transfer learning for both feature extraction and classification. To mitigate the inter-patient seizure pattern variation, we adapt Meta UPdate Strategy (MUPS) for four-class classification on the world's largest public seizure dataset of EEGs, Temple University Seizure Corpus (TUSZ). Different from existing works on binary seizure detection, we use the non-seizure samples and the top three most frequent seizure types for seizure detection. Our experiments show that the meta transfer learning approach achieves macro-F1 of 0.5103 and AUC of 0.6792, which outperforms the baseline learners (shallow and deep) by mitigating patient-to patient variations. We demonstrate the effectiveness of meta transfer learning in feature extraction and classification for multi-class seizure detection.

Seizure detection is a major goal for simplifying the workflow of clinicians working on EEG records. Current algorithms can only detect seizures effectively for patients already presented to the classifier. These algorithms are hard to generalize outside the initial training set without proper regularization and fail to capture seizures from the larger population. We proposed a data processing pipeline for seizure detection on an intra-patient dataset from the world's largest public EEG seizure corpus. We created spatially and session invariant features by forcing our networks to rely less on exact combinations of channels and signal amplitudes, but instead to learn dependencies towards seizure detection. For comparison, the baseline results without any additional regularization on a deep learning model achieved an F1 score of 0.544. By using random rearrangements of channels on each minibatch to force the network to generalize to other combinations of channels, we increased the F1 score to 0.629. By using random rescale of the data within a small range, we further increased the F1 score to 0.651 for our best model. Additionally, we applied adversarial multi-task learning and achieved similar results. We observed that session and patient specific dependencies were causing overfitting of deep neural networks, and the most overfitting models learnt features specific only to the EEG data presented. Thus, we created networks with regularization that the deep learning did not learn patient and session-specific features. We are the first to use random rearrangement, random rescale, and adversarial multitask learning to regularize intra-patient seizure detection and have increased sensitivity to 0.86 comparing to baseline study.

Cyber-physical systems (CPS) are controlling many critical and sensitive aspects of our physical world while being continuously exposed to potential cyber-attacks. These systems typically have limited performance, memory, and energy reserves, which limits their ability to run existing advanced malware protection, and that, in turn, makes securing them very challenging. To tackle these problems, this paper proposes, REMOTE, a new robust framework to detect malware by externally observing Electromagnetic (EM) signals emitted by an electronic computing device (e.g., a microprocessor) while running a known application, in real-time and with a low detection latency, and without any a priori knowledge of the malware. REMOTE does not require any resources or infrastructure on, or any modifications to, the monitored system itself, which makes REMOTE especially suitable for malware detection on resource-constrained devices such as embedded devices, CPSs, and Internet of Things (IoT) devices where hardware and energy resources may be limited. To demonstrate the usability of REMOTE in real-world scenarios, we port two real-world programs (an embedded medical device and an industrial PID controller), each with a meaningful attack (a code-reuse and a code-injection attack), to four different hardware platforms. We also port shellcode-based DDoS and Ransomware attacks to five different standard applications on an embedded system. To further demonstrate the applicability of REMOTE to commercial CPS, we use REMOTE to monitor a Robotic Arm. Our results on all these different hardware platforms show that, for all attacks on each of the platforms, REMOTE successfully detects each instance of an attack and has99.9 percent true positive rates) under all these conditions. We also compare REMOTE to prior work EDDIE [1] and SYNDROME [2], and demonstrate that these prior work are unable to achieve high accuracy under these variations.