List of relevant projects
AI in the pediatric intensive care unit (started 15th Nov. 2021 30th Dec.2023): We focused on predicting organ dysfunction using PICU data, which included vital signs such as heart rate, laboratory values like the number of leukocytes in the blood, and patient information such as age.We conducted multiple rounds of data analysis, visualization, and maintenance on the dataset in collaboration with medical doctors to ensure high-quality clinical data. Subsequently, we trained machine learning models to make predictions. Throughout the project, we faced several challenges, including high levels of missing data, imbalanced datasets, limited data availability, and the black-box nature of certain models. To address these issues, we employed various techniques, such as data imputation, masking, explainable algorithms, and incorporating additional patient data.
Explainable AI for neural time series black box models (started 1 5th Jan. 202 3 20th Oct . 2023): We introduce a framework for explaining decisions of black-box time series models. Our proposed framework utilizes a parameterized model for mask generation. Its effectiveness is demonstrated through experiments over real-world datasets and time series classification models. Our framework is also computationally efficient and can be deployed with any time series classification model.
Additionally, I contributed to developing deep learning models for automated heart sound analysis. Our proposed Deep Neural Network (DNN) architecture was adapted from the well-known Convolutional Neural Network (CNN) and ResNet architectures. The project was implemented using PyTorch. Github: https://github.com/Seham-Nasr/E2EModel-classifier2022