BME_lecture2

What is NOT one of the challenges in AI medical images?

• Doctors can easily make annotation for large scale dataset

• Noisy labels from multiple experts

• Privacy and legal issues

• Class imbalance

What is NOT the classic medical imaging problem?

• Registration

• Classification

• Segmentation

• Federated learning

Example of uni-modal registration?

• Cardiac MRI

• T1w-T2w MRI

• CT-MRI

• MRI-Ultrasound

One of the SOTA networks for medical image segmentation?

• SE-ResNet

• U-Net (and its variants)

• Inception-Net

• Mobile-Net

A technique to transfer knowledge from pre-trained model (from large scale natural images) to be used for medical images

• Curriculum learning

• Federated learning

• Transfer learning

• Multi-modal learning

What is the benefit of domain adaptation in medical images?

• Increase the domain shift

• Reducing the domain generalization

• Transfer knowledge from pre-trained model from large scale natural images

• Improve the performance of a target model with insufficient or lack of annotated data

Which one does NOT require manually annotated labels?

• Ensemble learning

• Supervised learning

• Self-supervised learning

• Multi-modal learning

What is PACS?

• picture archiving and communications systems

• picture asynchronous constellation systems

• patient archive and communication system

• patient as synced and computer system

What is the benefit of federated learning?

• Send all patients details to the central system

• Share the privacy protocols

• Train the data without actually sending data to the central node.

• Heavy computing process

which one is not biomedical image?

• Whole Slide Images

• Fundus image

• SPECT image

• Depth image