Projects and Publications

You can also find my publications on my Google Scholar profile.

2024 Pediatric Sepsis Data Challenge

Published in Pediatric Critical Care Medicine, 2024

This competition aims to facilitate the development of open-source algorithms to predict in-hospital mortality in children with sepsis. The challenge is to first develop an algorithm using a synthetic training dataset, which will then be scored according to standard diagnostic testing criteria, and then be evaluated against a nonsynthetic test dataset. The datasets originate from admissions to six hospitals in Uganda (2017–2020) and include 3837 children, 6 to 60 months old, who were confirmed or suspected to have a diagnosis of sepsis. The synthetic dataset was created from a random subset of the original data. The test validation dataset closely resembles the synthetic dataset. The challenge should generate an optimal model for predicting in-hospital mortality. Following external validation, this model could be used to improve the outcomes for children with proven or suspected sepsis in low- and middle-income settings.
Paper Challenge Website

Social Media as a Sensor: Analyzing Twitter Data for Breast Cancer Medication Effects Using Natural Language Processing

Published in Artificial Intelligence in Medicine (AIME), 2024

Breast cancer is a significant public health concern and is the leading cause of cancer-related deaths among women. Despite advances in breast cancer treatments, medication non-adherence remains a major problem. As electronic health records do not typically capture patient-reported outcomes that may reveal information about medication-related experiences, social media presents an attractive resource for enhancing our understanding of the patients’ treatment experiences. In this paper, we developed natural language processing (NLP) based methodologies to study information posted by an automatically curated breast cancer cohort from social media. We employed a transformer-based classifier to identify breast cancer patients/survivors on X (Twitter) based on their self-reported information, and we collected longitudinal data from their profiles. We then designed a multi-layer rule-based model to develop a breast cancer therapy-associated side effect lexicon and detect patterns of medication usage and associated side effects among breast cancer patients. 1,454,637 posts were available from 583,962 unique users, of which 62,042 (10.6%) were detected as breast cancer members using our transformer-based model. 198 cohort members mentioned breast cancer medications, with tamoxifen as the most common. Our side effect lexicon identified well-known side effects of hormone and chemotherapy. Furthermore, it discovered a subjective feeling towards cancer and medications, which may suggest a pre-clinical phase of side effects or emotional distress. This analysis highlighted not only the utility of NLP techniques in unstructured social media data to identify self-reported breast cancer posts, medication usage patterns, and treatment side effects but also the richness of social data to answer such clinical questions.
Paper Code

Improving mixed-integer temporal modeling by generating synthetic data using conditional generative adversarial networks: A case study of fluid overload prediction in the intensive care unit

Published in Computers in Biology and Medicine, 2024

The challenge of mixed-integer temporal data, which is particularly prominent for medication use in the critically ill, limits the performance of predictive models. The purpose of this evaluation was to pilot test integrating synthetic data within an existing dataset of complex medication data to improve machine learning model prediction of fluid overload. Materials and methods This retrospective cohort study evaluated patients admitted to an ICU ≥ 72 h. Four machine learning algorithms to predict fluid overload after 48–72 h of ICU admission were developed using the original dataset. Then, two distinct synthetic data generation methodologies (synthetic minority over-sampling technique (SMOTE) and conditional tabular generative adversarial network (CTGAN)) were used to create synthetic data. Finally, a stacking ensemble technique designed to train a meta-learner was established. Models underwent training in three scenarios of varying qualities and quantities of datasets. Results Training machine learning algorithms on the combined synthetic and original dataset overall increased the performance of the predictive models compared to training on the original dataset. The highest performing model was the meta-model trained on the combined dataset with 0.83 AUROC while it managed to significantly enhance the sensitivity across different training scenarios. Discussion The integration of synthetically generated data is the first time such methods have been applied to ICU medication data and offers a promising solution to enhance the performance of machine learning models for fluid overload, which may be translated to other ICU outcomes. A meta-learner was able to make a trade-off between different performance metrics and improve the ability to identify the minority class.
Paper Code

Development and validation of a model for endotracheal intubation and mechanical ventilation prediction in PICU patients

Published in Pediatric Critical Care Medicine, 2024

OBJECTIVES: To develop and externally validate an intubation prediction model for children admitted to a PICU using objective and routinely available data from the electronic medical records (EMRs). DESIGN: Retrospective observational cohort study. SETTING: Two PICUs within the same healthcare system: an academic, quaternary care center (36 beds) and a community, tertiary care center (56 beds). PATIENTS: Children younger than 18 years old admitted to a PICU between 2010 and 2022. MEASUREMENTS AND MAIN RESULTS: Clinical data was extracted from the EMR. PICU stays with at least one mechanical ventilation event (≥ 24 hr) occurring within a window of 1–7 days after hospital admission were included in the study. Of 13,208 PICU stays in the derivation PICU cohort, 1,175 (8.90%) had an intubation event. In the validation cohort, there were 1,165 of 17,841 stays (6.53%) with an intubation event. We trained a Categorical Boosting (CatBoost) model using vital signs, laboratory tests, demographic data, medications, organ dysfunction scores, and other patient characteristics to predict the need of intubation and mechanical ventilation using a 24-hour window of data within their hospital stay. We compared the CatBoost model to an extreme gradient boost, random forest, and a logistic regression model. The area under the receiving operating characteristic curve for the derivation cohort and the validation cohort was 0.88 (95% CI, 0.88–0.89) and 0.92 (95% CI, 0.91–0.92), respectively. CONCLUSIONS: We developed and externally validated an interpretable machine learning prediction model that improves on conventional clinical criteria to predict the need for intubation in children hospitalized in a PICU using information readily available in the EMR. Implementation of our model may help clinicians optimize the timing of endotracheal intubation and better allocate respiratory and nursing staff to care for mechanically ventilated children.
Paper

Meta-learning in healthcare: A survey

Published in arXiv, 2023

As a subset of machine learning, meta-learning, or learning to learn, aims at improving the model’s capabilities by employing prior knowledge and experience. A meta-learning paradigm can appropriately tackle the conventional challenges of traditional learning approaches, such as insufficient number of samples, domain shifts, and generalization. These unique characteristics position meta-learning as a suitable choice for developing influential solutions in various healthcare contexts, where the available data is often insufficient, and the data collection methodologies are different. This survey discusses meta-learning broad applications in the healthcare domain to provide insight into how and where it can address critical healthcare challenges. We first describe the theoretical foundations and pivotal methods of meta-learning. We then divide the employed meta-learning approaches in the healthcare domain into two main categories of multi/single-task learning and many/few-shot learning and survey the studies. Finally, we highlight the current challenges in meta-learning research, discuss the potential solutions and provide future perspectives on meta-learning in healthcare.
Paper

Pharmacophenotype identification of intensive care unit medications using unsupervised cluster analysis of the ICURx common data model

Published in Critical Care, 2023

Identifying patterns within ICU medication regimens may help artificial intelligence algorithms to better predict patient outcomes; however, machine learning methods incorporating medications require further development, including standardized terminology. The Common Data Model for Intensive Care Unit (ICU) Medications (CDM-ICURx) may provide important infrastructure to clinicians and researchers to support artificial intelligence analysis of medication-related outcomes and healthcare costs. Using an unsupervised cluster analysis approach in combination with this common data model, the objective of this evaluation was to identify novel patterns of medication clusters (termed ‘pharmacophenotypes’) correlated with ICU adverse events (e.g., fluid overload) and patient-centered outcomes (e.g., mortality). This was a retrospective, observational cohort study of 991 critically ill adults. To identify pharmacophenotypes, unsupervised machine learning analysis with automated feature learning using restricted Boltzmann machine and hierarchical clustering was performed on the medication administration records of each patient during the first 24 h of their ICU stay. Hierarchical agglomerative clustering was applied to identify unique patient clusters. Distributions of medications across pharmacophenotypes were described, and differences among patient clusters were compared using signed rank tests and Fisher’s exact tests, as appropriate. A total of 30,550 medication orders for the 991 patients were analyzed; five unique patient clusters and six unique pharmacophenotypes were identified. For patient outcomes, compared to patients in Clusters 1 and 3, patients in Cluster 5 had a significantly shorter duration of mechanical ventilation and ICU length of stay (p < 0.05); for medications, Cluster 5 had a higher distribution of Pharmacophenotype 1 and a smaller distribution of Pharmacophenotype 2, compared to Clusters 1 and 3. For outcomes, patients in Cluster 2, despite having the highest severity of illness and greatest medication regimen complexity, had the lowest overall mortality; for medications, Cluster 2 also had a comparably higher distribution of Pharmacophenotype 6. The results of this evaluation suggest that patterns among patient clusters and medication regimens may be observed using empiric methods of unsupervised machine learning in combination with a common data model. These results have potential because while phenotyping approaches have been used to classify heterogenous syndromes in critical illness to better define treatment response, the entire medication administration record has not been incorporated in those analyses. Applying knowledge of these patterns at the bedside requires further algorithm development and clinical application but may have the future potential to be leveraged in guiding medication-related decision making to improve treatment outcomes.
Paper Code

Automated major depressive disorder classification using deep convolutional neural networks and Choquet fuzzy integral fusion

Published in 2022 IEEE Symposium Series on Computational Intelligence (SSCI), 2022

Major depressive disorder (MDD) is a common and severe ailment impacting functional frailty, while its concrete manifestations have been shrouded in mystery. Hence, manual diagnosis of MDD is an arduous and subjective task. Despite the aid of electroencephalogram (EEG) signals in the detection, developing intelligent systems are required to improve clinical utility, performance, and efficiency. In this study, we focus on the automated detection of MDD via raw EEG data using convolutional neural networks (CNN). For this objective, we first extracted the short-time Fourier transform (STFT) of EEG records for five distinct band powers and created an image representing the frequency oscillation of every channel during a resting state. Afterward, we applied three approaches to determine whether a subject is MDD or a Healthy individual. In the first approach, a 2D-CNN model was developed for each band power to detect MDD separately. Second, the outcomes of the developed models were used to establish a Choquet fuzzy integral fusion to classify subjects using all of the previous models. The third approach was dedicated to introducing a 3D-CNN architecture. This model received three-dimensional data by putting different band powers’ images together. The two last approaches achieved a 95.65% accuracy and 100% sensitivity to detect MDD. The proposed approaches can help clinicians as straightforward, efficient, and intelligent diagnostic tools for detecting MDD.
Paper

EISI: Extended inter-spike interval for mental health patients clustering based on mental health services and medications utilisation

Published in Medical Engineering & Physics, 2022

Mental health is vital in all human life stages, and managing mental healthcare service resources is crucial for providers. This paper presents a new method, called Extended Inter-Spike Interval (EISI), on identifying the patients with a similar utilisation of mental health services and medications. The EISI measures the distance between the utilisation patterns of the patients. Then, the pairwise distances are given to a developed split-and-merge Partitioning Around Medoids (PAM) clustering algorithm to identify the patients with similar utilisation patterns. To evaluate the proposed method, we use two years (2013–2014) of the 10% publicly available sample of the Australian Medicare Benefits Schedule (MBS) and Pharmaceutical Benefits Scheme (PBS) administrative data. Results show that mental health patients can be grouped into ten clusters with distinct and interpretable utilisations patterns. The largest cluster comprises individuals who only visit general practitioners and take psycholeptics medications for a short time. The smallest group contains occasional visits with general practitioners and regularly utilises psycholeptics and psychoanaleptics medications over long periods. The proposed method provides insights on whom to target and how to structure services for different groups of individuals with mental health conditions.
Paper

Automated detection of major depressive disorder with EEG signals: a time series classification using deep learning

Published in IEEE Access, 2022

Major depressive disorder (MDD) has been considered a severe and common ailment with effects on functional frailty, while its clear manifestations are shrouded in mystery. Hence, manual detection of MDD is a challenging and subjective task. Although Electroencephalogram (EEG) signals have shown promise in aiding diagnosis, further enhancement is required to improve accuracy, clinical utility, and efficiency. This study focuses on the automated detection of MDD using EEG data and deep neural network architecture. For this aim, first, a customized InceptionTime model is recruited to detect MDD individuals via 19-channel raw EEG signals. Then a channel-selection strategy, which comprises three channel-selection steps, is conducted to omit redundant channels. The proposed method achieved 91.67% accuracy using the full set of channels and 87.5% after channel reduction. Our analysis shows that i) only the first minute of EEG recording is sufficient for MDD detection, ii) models based on EEG recorded in eyes-closed resting-state outperform eyes-open conditions, and iii) customizing the InceptionTime model can improve its efficiency for different assignments. The proposed method is able to help clinicians as an efficient, straightforward, and intelligent diagnostic tool for the objective detection of MDD.
Paper Code

Pedestrian collision avoidance using deep reinforcement learning

Published in International journal of automotive technology, 2022

The use of intelligent systems to prevent accidents and safety enhancement in vehicles is becoming a requirement. Besides, the development of autonomous cars is progressing every day. One of the main challenges in transportation is the high mortality rate of vehicles colliding with pedestrians. This issue becomes severe due to various and abnormal situations. This paper proposes a new intelligent algorithm for pedestrian collision avoidance based on deep reinforcement learning. A deep Q-network (DQN) is designed to discover an optimal driving policy for pedestrian collision avoidance in diverse environments and conditions. The algorithm interacts with the vehicle and the pedestrian agents and uses a specific reward function to train the model. We have used Car Learning to Act (CARLA), an open-source autonomous driving simulator, for training and verifying the model in various conditions. Applying the proposed algorithm to a simulated environment reduces vehicles and pedestrians’ collision by about 64 %, depending on the environment. Our findings offer an early-warning solution to mitigate the risk of a crash of vehicles and pedestrians in the real world.
Paper

FCOD: Fast COVID-19 Detector based on deep learning techniques

Published in Informatics in medicine unlocked, 2021

The sudden COVID-19 pandemic has caused a serious global concern due to infections and mortality rates. It is a hazardous disease that has recently become the biggest crisis in the modern era. Due to the limitation of test kits and the need for screening and rapid diagnosis of patients, it is essential to perform a self-operating detection model as a fast recognition system to detect COVID-19 infection and prevent the spread among the people. In this paper, we propose a novel technique called Fast COVID-19 Detector (FCOD) to have a fast detection of COVID-19 using X-ray images. The FCOD is a deep learning model based on the Inception architecture that uses 17 depthwise separable convolution layers to detect COVID-19. Depthwise separable convolution layers decrease the computation costs, time, and they can have a reducing role in the number of parameters compared to the standard convolution layers. To evaluate FCOD, we used covid-chestxray-dataset, which contains 940 publicly available typical chest X-ray images. Our results show that FCOD can provide accuracy, F1-score, and AUC of 96%, 96%, and 0.95%, respectively in classifying COVID-19 during 0.014 s for each case. The proposed model can be employed as a supportive decision-making system to assist radiologists in clinics and hospitals to screen patients immediately.
Paper

SSP: Early prediction of sepsis using fully connected LSTM-CNN model

Published in Computers in biology and medicine, 2021

Sepsis is a life-threatening condition that occurs due to the body’s reaction to infections, and it is a leading cause of morbidity and mortality in hospitals. Early prediction of sepsis onset facilitates early interventions that promote the survival of suspected patients. However, reliable and intelligent systems for predicting sepsis are scarce. This paper presents a novel technique called Smart Sepsis Predictor (SSP) to predict sepsis onset in patients admitted to an intensive care unit (ICU). SSP is a deep neural network architecture that encompasses long short-term memory (LSTM), convolutional, and fully connected layers to achieve early prediction of sepsis. SSP can work in two modes; Mode 1 uses demographic data and vital signs, and Mode 2 uses laboratory test results in addition to demographic data and vital signs. To evaluate SSP, we have used the 2019 PhysioNet/CinC Challenge dataset, which includes the records of 40,366 patients admitted to the ICU. To compare SSP with existing state-of-the-art methods, we have measured the accuracy of the SSP in 4-, 8-, and 12-h prediction windows using publicly available data. Our results show that the SSP performance in Mode 1 and Mode 2 is much higher than existing methods, achieving an area under the receiver operating characteristic curve (AUROC) of 0.89 and 0.92, 0.88 and 0.87, and 0.86 and 0.84 for 4 h, 8 h, and 12 h before sepsis onset, respectively. Using ICU data, sepsis onset can be predicted up to 12 h in advance. Our findings offer an early solution for mitigating the risk of sepsis onset.
Paper

Alireza Rafiei

Projects and Publications