BE 645: Artificial Intelligence (AI) and Radiomics - Lecture 09 Recap (AI-Generated Podcast)

BE 645: Artificial Intelligence (AI) and Radiomics (Spring 2025)

The ninth lecture focuses on a radiomics workflow integrated with artificial intelligence, referencing the MedMNIST+ database. The workflow begins with acquiring and segmenting radiological images, using the AdrenalMNIST3D dataset as an example. AdrenalMNIST3D consists of 3D shape masks of adrenal glands, labeled by expert endocrinologists, and split into training, validation, and test sets. After image acquisition and segmentation, the process advances to extracting radiomics features from the regions of interest, followed by preprocessing to select the most significant features. Subsequently, the radiomics workflow involves developing, training, validating, and testing AI models for diagnosis prediction. The lecture lists normalization techniques and ML models. The radiomics workflow concludes with reporting performance metrics and focusing on explainability. The lecture also emphasizes the use of feature selection techniques (filter, wrapper, and embedded methods) to enhance model performance. Filter methods use statistical tests like the Chi-square test, mutual information, and ANOVA F-value. Wrapper methods evaluate feature usefulness by training models on different subsets, including forward selection, backward elimination, and recursive feature elimination. Embedded methods perform feature selection during model training, including Lasso (L1 regularization) and decision tree-based methods. The MedMNIST+ database is noted as a resource for standardized 2D and 3D biomedical image classification datasets.

Codes can be accessed using: https://github.com/HossamBalaha/BE-64...

Playlist from Spring 2025:    • BE 645: Artificial Intelligence (AI) and R...  

Old Playlist from Summer 2024:    • BE 645: Artificial Intelligence (AI) and R...  

Disclaimer: This is an AI-generated audio created using NotebookLM. The content may not reflect the opinions or expertise of a human author and is provided for informational or illustrative purposes only.