Using Artificial Intelligence to Interpret Clinical Flow Cytometry Datasets for Automated Disease Diagnosis and/or Monitoring.

Flow cytometry (FC) is routinely used for hematological disease diagnosis and monitoring. Advancement in this technology allows us to measure an increasing number of markers simultaneously, generating complex high-dimensional datasets. However, current analytic software and methods rely on experienced analysts to perform labor-intensive manual inspection and interpretation on a series of 2-dimensional plots via a complex, sequential gating process. With an aggravating shortage of professionals and growing demands, it is very challenging to provide the FC analysis results in a fast, accurate, and reproducible way. Artificial intelligence has been widely used in many sectors to develop automated detection or classification tools. Here we describe a type of machine learning method for developing automated disease classification and residual disease monitoring on clinical flow datasets.

Lung Cancer Prediction Using Electronic Claims Records: A Transformer-Based Approach.

Electronic claims records (ECRs) are large scale and longitudinal collections of individual's medical service seeking actions. Compared to in-hospital medical records (EMRs), ECRs are more standardized and cross-sites. Recently, there has been studies showing promising results on modeling claims data for a wide range of medical applications. However, few of them address the exclusion criteria on cohort selection to extract new incidence without prior signs and also often lack of emphasis on predicting cancer in early stages. In this work, we aim to design a lung cancer prediction framework using ECRs with rigorous exclusion design using state-of-the-art sequence-based transformer. Furthermore, this work presents one of the first results by applying disease prediction model to the entire population in Taiwan. The result shows over 2.1 predictive power, 5 average positive predictive value (PPV), and 0.668 area under curve (AUC) in all-stage lung cancer and around 2.0 predictive power, 1 average PPV and 0.645 AUC in early-stage in our dataset. Sub-cohort analysis could funnel high precision selective group into prioritized clinical examination. Onset analysis validates the effect of our exclusion criteria. This work presents comprehensive analyses on lung cancer prediction, and the proposed approach can serve as a state-of-the-art disease risk prediction framework on claims data.

A Coarse-to-Fine Pathology Patch Selection for Improving Gene Mutation Prediction in Acute Myeloid Leukemia.

Identifying gene mutation is essential to prognosis and therapeutic decisions for acute myeloid leukemia (AML) but the current gene analysis is inefficient and non-scalable. Pathological images are readily accessible and can be effectively modeled using deep learning. This work aims at predicting gene mutation directly by modeling bone marrow smear images. Traditionally, bone marrow smear slides are cropped into patches with manual segmentation for patch-level modeling. Slide-level modeling, such as multi-instance learning, could aggregate patches for holistic modeling, though suffer from excessive redundancy. In this study, we propose a discriminative multi-instance approach to select useful patches in a coarse-to-fine process. Specifically, we preprocess a slide into patches by using a trained pre-selector network. Then, we rule out low quality patches in the coarse selection with known prior knowledge, and refine the model using gene-discriminative patches in the fine selection. We evaluate the framework for CEBPA, FLT3, and NPM1 gene mutation prediction and obtain 71.67%, 56.26%, and 56.34% F1-score. Further analysis show the effect of different selection criteria on prediction gene mutations using pathological images.

A Chunking-for-Pooling Strategy for Cytometric Representation Learning for Automatic Hematologic Malignancy Classification.

Differentiating types of hematologic malignancies is vital to determine therapeutic strategies for the newly diagnosed patients. Flow cytometry (FC) can be used as diagnostic indicator by measuring the multi-parameter fluorescent markers on thousands of antibody-bound cells, but the manual interpretation of large scale flow cytometry data has long been a time-consuming and complicated task for hematologists and laboratory professionals. Past studies have led to the development of representation learning algorithms to perform sample-level automatic classification. In this work, we propose a chunking-for-pooling strategy to include large-scale FC data into a supervised deep representation learning procedure for automatic hematologic malignancy classification. The use of discriminatively-trained representation learning strategy and the fixed-size chunking and pooling design are key components of this framework. It improves the discriminative power of the FC sample-level embedding and simultaneously addresses the robustness issue due to an inevitable use of down-sampling in conventional distribution based approaches for deriving FC representation. We evaluated our framework on two datasets. Our framework outperformed other baseline methods and achieved 92.3% unweighted average recall (UAR) for four-class recognition on the UPMC dataset and 85.0% UAR for five-class recognition on the hema.to dataset. We further compared the robustness of our proposed framework with that of the traditional downsampling approach. Analysis of the effects of the chunk size and the error cases revealed further insights about different hematologic malignancy characteristics in the FC data.

A Machine Learning Approach to the Classification of Acute Leukemias and Distinction From Nonneoplastic Cytopenias Using Flow Cytometry Data.

OBJECTIVES: Flow cytometry (FC) is critical for the diagnosis and monitoring of hematologic malignancies. Machine learning (ML) methods rapidly classify multidimensional data and should dramatically improve the efficiency of FC data analysis. We aimed to build a model to classify acute leukemias, including acute promyelocytic leukemia (APL), and distinguish them from nonneoplastic cytopenias. We also sought to illustrate a method to identify key FC parameters that contribute to the model's performance. METHODS: Using data from 531 patients who underwent evaluation for cytopenias and/or acute leukemia, we developed an ML model to rapidly distinguish among APL, acute myeloid leukemia/not APL, acute lymphoblastic leukemia, and nonneoplastic cytopenias. Unsupervised learning using gaussian mixture model and Fisher kernel methods were applied to FC listmode data, followed by supervised support vector machine classification. RESULTS: High accuracy (ACC, 94.2%; area under the curve [AUC], 99.5%) was achieved based on the 37-parameter FC panel. Using only 3 parameters, however, yielded similar performance (ACC, 91.7%; AUC, 98.3%) and highlighted the significant contribution of light scatter properties. CONCLUSIONS: Our findings underscore the potential for ML to automatically identify and prioritize FC specimens that have critical results, including APL and other acute leukemias.

Federated Learning via Conditional Mutual Learning for Alzheimer's Disease Classification on T1w MRI.

Data-driven deep learning has been considered a promising method for building powerful models for medical data, which often requires a large amount of diverse data to be sufficiently effective. However, the expensive cost of collecting and the privacy constraints lead to the fact that existing medical datasets are small-scale and distributed. Federated learning via model distillation is a data-private collaborative learning where the model can leverage all available data without direct sharing. The data knowledge is shared by distillation through the multi-site average prediction scores on the public dataset. However, the average consensus is suboptimal to individual client due to data domain shift in MRI data caused by acquisition protocols, recruitment criteria, etc. In this work, we propose a federated conditional mutual learning (FedCM) to improve the performance by considering the clients' local performance and the similarity between clients. This work is the first federated learning on multi-dataset Alzheimer's disease classification by 3DCNN using T1w MRI. Our method achieves the best recognition rates comparing with FedMD and other frameworks. Further visualization and relevance ranking on the region of interests (ROI) in human brains implies that the left hemisphere may have greater relevance than the right hemisphere does. Several potential regions are listed for future investigation.

Empowering the public during the COVID pandemic through interactive social media platform.

During the COVID-19 pandemic, healthcare professionals and academic facilities are called to provide leadership in disseminating accurate and timely information through approaches that meet the needs of the public. Graduate students from a university in Taiwan collaborated with experts to provide interactive live broadcasting sessions on the COVID-related topics to the public through the Facebook platform. The broadcasting sessions also trained the students to communicate COVID-related information through succinct and interactive presentations. Twelve broadcasting sessions were conducted twice a week for three weeks in May 2020. Upon completion of the broadcasting sessions, students demonstrated growth in professional confidence, assessment of the public's knowledge gaps and needs, and preparation and delivery of professional live broadcasts. We recommend creating a live broadcast training application through an artificial intelligence (AI) expert system. Multidisciplinary academic-practice collaboration in preparing for the broadcasting and engaging in dialogues with the public is recommended.

Learning a Phenotypic-Attribute Attentional Brain Connectivity Embedding for ADHD Classification using rs-fMRI.

Automated diagnosis of Attention Deficit/Hyperactivity Disorder (ADHD) from brain's functional imaging has gained more interest due to its high prevalence rates among children. While phenotypic information, such as age and gender, is known to be important in diagnosing ADHD and critically affects the representation derived from fMRI brain images, limited studies have integrated phenotypic information when learning discriminative embedding from brain imaging for such an automatic classification task. In this work, we propose to integrate age and gender attributes through attention mechanism that is jointly optimized when learning a brain connectivity embedding using convolutional variational autoencoder derived from resting state functional magnetic resonance imaging (rs-fMRI) data. Our proposed framework achieves a state-of-the-art average of 86.22% accuracy in ADHD vs. typical develop control (TDC) binary classification task evaluated across five public ADHD-200 competition datasets. Furthermore, our analysis points out that there are insufficient linked connections to the brain region of precuneus in the ADHD group.

Behavior Score-Embedded Brain Encoder Network for Improved Classification of Alzheimer Disease Using Resting State fMRI.

The ability to accurately detect onset of dementia is important in the treatment of the disease. Clinically, the diagnosis of Alzheimer Disease (AD) and Mild Cognitive Impairment (MCI) patients are based on an integrated assessment of psychological tests and brain imaging such as positron emission tomography (PET) and anatomical magnetic resonance imaging (MRI). In this work using two different datasets, we propose a behavior score-embedded encoder network (BSEN) that integrates regularly adminstrated psychological tests information into the encoding procedure of representing subject's resting-state fMRI data for automatic classification tasks. BSEN is based on a 3D convolutional autoencoder structure with contrastive loss jointly optimized using behavior scores from Mini-Mental State Examination (MMSE) and Clinical Dementia Rating (CDR). Our proposed classification framework of using BSEN achieved an overall recognition accuracy of 59.44% (3-class classification: AD, MCI and Healthy Control), and we further extracted the most discriminative regions between healthy control (HC) and AD patients.

A Knowledge-Reserved Distillation with Complementary Transfer for Automated FC-based Classification Across Hematological Malignancies.

Acute leukemia often comes with life-threatening prognosis outcome and remains a critical clinical issue today. The implementation of measurable residual disease (MRD) using flow cytometry (FC) is highly effective but the interpretation is time-consuming and suffers from physician idiosyncrasy. Recent machine learning algorithms have been proposed to automatically classify acute leukemia samples with and without MRD to address this clinical need. However, most prior works either validate only on a small data cohort or focus on one specific type of leukemia which lacks generalization. In this work, we propose a transfer learning approach in performing automatic MRD classification that takes advantage of a large scale acute myeloid leukemia (AML) database to facilitate better learning on a small cohort of acute lymphoblastic leukemia (ALL). Specifically, we develop a knowledge-reserved distilled AML pre-trained network with ALL complementary learning to enhance the ALL MRD classification. Our framework achieves 84.5% averaged AUC which shows its transferability across acute leukemia, and our further analysis reveals that younger and elder ALL patient samples benefit more from using the pre-trained AML model.

Deriving A Novel Health Index Using A Large-Scale Population Based Electronic Health Record With Deep Networks.

Health indexes are useful tools for monitoring the health condition of a population and can be used to guide healthcare policy of governments. However, most health indexes are constructed by using statistical methods to summarize recent adverse events (e.g., mortality). Information from these tools may reflect merely the impact of prior health policy holistically and can hardly indicate the most recent dynamics and its impact on future health conditions. As the advancements in medications and medical techniques rapidly evolve, there is a need of new health indexes that can reflect the most recent predictive health condition of a population and can easily be summarized with respect of any sub-population of interest. In this work, we develop a novel health index by using deep learning technique on a large-scale and longitudinal population based electronic health record (EHR). Three deep neural network (DNN) models were trained to predict 4-year event rates of mortality, hospitalization and cancer occurrence at an individual-level. Platt calibration approach was used to transform DNN output scores into estimated event risks. A novel health index is then constructed by weighted scoring these calibrated event risks. This individual-level health index not only provide a better predictive power but can also be flexibly summarized for different regions or sub-populations of interest - hence providing objective insights to develop precise personal or national policy beyond conventional health index.

Investigation of functional brain network reconfiguration during vocal emotional processing using graph-theoretical analysis.

Vocal expression is essential for conveying the emotion during social interaction. Although vocal emotion has been explored in previous studies, little is known about how perception of different vocal emotional expressions modulates the functional brain network topology. In this study, we aimed to investigate the functional brain networks under different attributes of vocal emotion by graph-theoretical network analysis. Functional magnetic resonance imaging (fMRI) experiments were performed on 36 healthy participants. We utilized the Power-264 functional brain atlas to calculate the interregional functional connectivity (FC) from fMRI data under resting state and vocal stimuli at different arousal and valence levels. The orthogonal minimal spanning trees method was used for topological filtering. The paired-sample t-test with Bonferroni correction across all regions and arousal-valence levels were used for statistical comparisons. Our results show that brain network exhibits significantly altered network attributes at FC, nodal and global levels, especially under high-arousal or negative-valence vocal emotional stimuli. The alterations within/between well-known large-scale functional networks were also investigated. Through the present study, we have gained more insights into how comprehending emotional speech modulates brain networks. These findings may shed light on how the human brain processes emotional speech and how it distinguishes different emotional conditions.

Development of an intelligent decision support system for ischemic stroke risk assessment in a population-based electronic health record database.

BACKGROUND: Intelligent decision support systems (IDSS) have been applied to tasks of disease management. Deep neural networks (DNNs) are artificial intelligent techniques to achieve high modeling power. The application of DNNs to large-scale data for estimating stroke risk needs to be assessed and validated. This study aims to apply a DNN for deriving a stroke predictive model using a big electronic health record database. METHODS AND RESULTS: The Taiwan National Health Insurance Research Database was used to conduct a retrospective population-based study. The database was divided into one development dataset for model training (~70% of total patients for training and ~10% for parameter tuning) and two testing datasets (each ~10%). A total of 11,192,916 claim records from 840,487 patients were used. The primary outcome was defined as any ischemic stroke in inpatient records within 3 years after study enrollment. The DNN was evaluated using the area under the receiver operating characteristic curve (AUC or c-statistic). The development dataset included 672,214 patients (a total of 8,952,000 records) of whom 2,060 patients had stroke events. The mean age of the population was 35.5±20.2 years, with 48.5% men. The model achieved AUC values of 0.920 (95% confidence interval [CI], 0.908-0.932) in testing dataset 1 and 0.925 (95% CI, 0.914-0.937) in testing dataset 2. Under a high sensitivity operating point, the sensitivity and specificity were 92.5% and 79.8% for testing dataset 1; 91.8% and 79.9% for testing dataset 2. Under a high specificity operating point, the sensitivity and specificity were 80.3% and 87.5% for testing dataset 1; 83.7% and 87.5% for testing dataset 2. The DNN model maintained high predictability 5 years after being developed. The model achieved similar performance to other clinical risk assessment scores. CONCLUSIONS: Using a DNN algorithm on this large electronic health record database is capable of obtaining a high performing model for assessment of ischemic stroke risk. Further research is needed to determine whether such a DNN-based IDSS could lead to an improvement in clinical practice.

Learning a Cytometric Deep Phenotype Embedding for Automatic Hematological Malignancies Classification.

Identification of minimal residual disease (MRD) is important in assessing the prognosis of acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS). The current best clinical practice relies heavily on Flow Cytometry (FC) examination. However, the current FC diagnostic examination requires trained physicians to perform lengthy manual interpretation on high-dimensional FC data measurements of each specimen. The difficulty in handling idiosyncrasy between interpreters along with the time-consuming diagnostic process has become one of the major bottlenecks in advancing the treatment of hematological diseases. In this work, we develop an automatic MRD classifications (AML, MDS, normal) algorithm based on learning a deep phenotype representation from a large cohort of retrospective clinical data with over 2000 real patients' FC samples. We propose to learn a cytometric deep embedding through cell-level autoencoder combined with specimen-level latent Fisher-scoring vectorization. Our method achieves an average AUC of 0.943 across four different hematological malignancies classification tasks, and our analysis further reveals that with only half of the FC markers would be sufficient in obtaining these high recognition accuracies.

Predicting Gastrointestinal Bleeding Events from Multimodal In-Hospital Electronic Health Records Using Deep Fusion Networks.

Applying machine learning (ML) methods on electronic health records (EHRs) that accurately predict the occurrence of a variety of diseases or complications related to medications can contribute to improve healthcare quality. EHRs by nature contain multiple modalities of clinical data from heterogeneous sources that require proper fusion strategy. The deep neural network (DNN) approach, which possesses the ability to learn classification and feature representation, is well-suited to be employed in this context. In this study, we collect a large in-hospital EHR database to develop analytics in predicting 1-year gastrointestinal (GI) bleeding hospitalizations for patients taking anticoagulants or antiplatelet drugs. A total of 815,499 records (16,757 unique patients) are used in this study with three different available EHR modalities (disease diagnoses, medications usage, and laboratory testing measurements). We compare the performances of 4 deep multimodal fusion models and other ML approaches. NNs result in higher prediction performances compare to random forest (RF), gradient boosting decision tree (GBDT), and logistic regression (LR) approaches. We further demonstrate that deep multimodal NNs with early fusion can obtain the best GI bleeding predictive power (area under the receiver operator curve [AUROC] 0.876), which is significantly better than the HAS-BLED score (AUROC 0.668).

A BLSTM with Attention Network for Predicting Acute Myeloid Leukemia Patient's Prognosis using Comprehensive Clinical Parameters.

The prognosis management is crucial for highrisk disease like Acute Myeloid Leukemia (AML) in order to support decisions of clinical treatment. However, the challenges of accurate and consistent forecasting lie in the high variability of the disease outcomes and the complexity of the multiple clinical measurements available over the course of the treatment. In order to capture the multi-dimensional and longitudinal aspect of these comprehensive clinical parameters, we utilize an attention-based bi-directional long shortterm memory (Att-BLSTM) network to predict AML patient's survival and relapse. Specifically, we gather a 10-year worth of real patient's clinical data including blood test, medication, HSCT status, and gene mutation information. Our proposed Att-BLSTM framework achieves 77.1% and 67.3% AUC in tasks of predicting the next 2-year mortality and disease relapse with these comprehensive clinical parameters, and our further analysis demonstrates that a next 0 to 3 months prediction performs equally well, i.e., 74.8% and 67% AUC for mortality and relapse respectively.

Improving Automatic Tremor and Movement Motor Disorder Severity Assessment for Parkinson's Disease with Deep Joint Training.

Parkinson's disease (PD) is one of the most severe and common disease globally. PD induces motor system impairment causing symptoms such as shaking, rigidity, slowness of movement, body tremor and difficulty with walking. Clinically, accurately and objectively assessing the severity of PD symptoms is critical in controlling appropriate dosage of Levodopa to prevent unwanted side effect of switching between Dyskinesia and PD. The unified Parkinson's disease rating scale published by the Movement Disorder Society (MDS-UPDRS) is an validated instrument regularly administrated by trained physician to assess the severity of a PD patient's motor disorder. In this work, we aim at advancing vision-based automatic motor disorder assessment, specifically hand tremor and movement, for PD patients during UPDRS. Our proposed method leverages information across the two behavior tasks simultaneously via deep joint training to improve each single task's, i.e., tremor and movement, severity classification rate. We evaluate our framework on a large cohort of 106 PD patients, and with our proposed deep joint training framework, we achieve accuracy of 78.01% and 80.60% in right and left hand movement binary classification; in terms of tremor severity classification, our approach obtains an enhanced recognition rates of 72.20% and 71.10% for right and left hand respectively.

Improving Young Stroke Prediction by Learning with Active Data Augmenter in a Large-Scale Electronic Medical Claims Database.

Electronic medical claims (EMC) database has been successfully used for predicting occurrences of stroke and a variety of other diseases. However, inadequate predictive performances have been observed in cases of rare occurrences due to both insufficient training samples and highly imbalanced class distribution. In this work, our aim is to improve stroke prediction, especially for young age group (25-45 year-old) in a large population-based EMC database (552,898 subjects). We learn a young stroke predictive deep neural network model using a novel active data augmenter. The augmenter selects the most informative EHR data samples from old age stroke patients. This approach achieves 9.3% and 8.2% area under the receiver operating characteristic curve (AUC) value improvements compared to training directly with only young age group data and training all age groups data, respectively. We further provide analyses on the AUC values obtained as a function of the training data size, and the amount and the type of augmented data samples.

Development of a rapid and economic in vivo electrocardiogram platform for cardiovascular drug assay and electrophysiology research in adult zebrafish.

Zebrafish is a popular and favorable model organism for cardiovascular research, with an increasing number of studies implementing functional assays in the adult stage. For example, the application of electrocardiography (ECG) in adult zebrafish has emerged as an important tool for cardiac pathophysiology, toxicity, and chemical screen studies. However, few laboratories are able to perform such functional analyses due to the high cost and limited availability of a convenient in vivo ECG recording system. In this study, an inexpensive ECG recording platform and operation protocol that has been optimized for adult zebrafish ECG research was developed. The core hardware includes integration of a ready-to-use portable ECG kit with a set of custom-made needle electrode probes. A combined anesthetic formula of MS-222 and isoflurane was first tested to determine the optimal assay conditions to minimize the interference to zebrafish cardiac physiology under sedation. For demonstration, we treated wild-type zebrafish with different pharmacological agents known to affect cardiac rhythms in humans. Conserved electrophysiological responses to these drugs were induced in adult zebrafish and recorded in real time. This economic ECG platform has the potential to facilitate teaching and training in cardiac electrophysiology with adult zebrafish and to promote future translational applications in cardiovascular medicine.