The Udaicty's AI for healthcare Nanodegree helps in learning skills to make clinical decisions using machine learning. My completion certificate of this program is here. In this program, I have learnt to:
- Recommend appropriate imaging modalities for common clinical applications of 2D medical imaging
- Perform exploratory data analysis (EDA) on 2D medical imaging data to inform model training and explain model performance
- Establish the appropriate ‘ground truth’ methodologies for training algorithms to label medical images
- Extract images from a DICOM dataset
- Train common CNN architectures to classify 2D medical images
- Translate outputs of medical imaging models for use by a clinician
- Plan necessary validations to prepare a medical imaging model for regulatory approval
- Detect major clinical abnormalities in a DICOM dataset
- Train machine learning models for classification tasks using real-world 3D medical imaging data
- Integrate models into a clinician’s workflow and troubleshoot deployments
- Build machine learning models in a manner that is compliant with U.S. healthcare data security and privacy standards
- Use the TensorFlow Dataset API to scalably extract, transform, and load datasets that are aggregated at the line, encounter, and longitudinal (patient) data levels
- Analyze EHR datasets to check for common issues (data leakage, statistical properties, missing values, high cardinality) by performing exploratory data analysis with TensorFlow Data Analysis and Validation library
- Create categorical features from Key Industry Code Sets (ICD, CPT, NDC) and reduce dimensionality for high cardinality features
- Use TensorFlow feature columns on both continuous and categorical input features to create derived features (bucketing, cross-features, embeddings)
- Use Shapley values to select features for a model and identify the marginal contribution for each selected feature
- Analyze and determine biases for a model for key demographic groups
- Use the TensorFlow Probability library to train a model that provides uncertainty range predictions in order to allow for risk adjustment/prioritization and triaging of predictions
- Preprocess data (eliminate “noise”) collected by IMU, PPG, and ECG sensors based on mechanical, physiology and environmental effects on the signal.
- Create an activity classification algorithm using signal processing and machine learning techniques
- Detect QRS complexes using one-dimensional time series processing techniques
- Evaluate algorithm performance without ground truth labels
- Generate a pulse rate algorithm that combines information from the PPG and IMU sensor streams