Please enable JavaScript.
Coggle requires JavaScript to display documents.
Medical Image Analysis (IEEE), Note: Datasets discussed are important,…
Medical Image Analysis (IEEE)
Electronic Health Records
Now being digitalized
Important records are medical images
Analyzed by radiologists
Slow
Prone to fatigue
Leads to inaccuracies
Takes time to develop experience
Takes time to train
Should be accurate
Errors will lead to harming patients
Use of machine learning (ML)
Faster
Could be more accurate
Challenges
Will it improve patient survival?
Patient willingness to use AI choices?
Increased in number
Techniques in Extraction and ML
Increase in image extraction
Modalities
Magnetic Resonance Imaging
Positron Emission Tomography
Computed Tomography
Retinal Photography
Histology Slides
Dermoscopy Images
Medical Image Analysis
Convolutional Neural Networks (CNN)
For image recognition tasks
Classification
Localization
Detection
Segmentation
Registration
Can handle 2D image analysis
Also 3D with some modifications
Analysis is dependent on datasets
Grand Challenges in Biomedical Image Analysis
Cancer Imaging Archive
ChestX-Ray 8
Machine Learning Architectures
Unsupervised Learning
Autoencoders
Feature representation
Dimension reduction
Generative Adversarial Networks
Generators
Discriminators
Used when there is only few data available
Supervised Learning
CNN
Fully Connected Layer
Pooling layer
Reduce number of parameters
Rectified Linear Unit Layer
Helps avoid vanishing gradients
Convolution layer
Intrinsic functions
Equivariant representation
Parameter sharing
sparse connection
Layers need not to be fully connected
Pass to high dimension layers
Capturing low-level features
Dependent on labelled data
Transfer Learning
Training on unrelated dataset
Transfer of trained parameters
Tuning of layers
Recurrent Neural Networks
Mainly used for segmentation
Used for analyzing sequential data
Applications
Classification
Usually binary
Can be multiclass
Example: determine if lung nodule or not
Metrics
Area Under Curve
Accuracy
Localization
Determine areas or regions where anomalies detected
May not be of interest for clinicians
Detection
Missing lesions
Detection of malignant cells
Segmentation
Analyze various parts of a particular organ
Mostly used for brains
Registration
Mostly used for surgeries
Implant placement
Note: Datasets discussed are important
Note: Top 20 papers are listed
Ker, J., Wang, L., Rao, J., and Lim, T. Deep
learning applications in medical image analysis. IEEE
Access 6 (2018), 9375–9389.