Vision and Med Tech

Ref - Link

Keep Exploring!!!

Day #196 - Structural Medical Image Analyses using Consistent Volume and Surface Image Processing

Key Summary

• 100 brain regions analysis
• Images -> Analysis -> Models

Image Processing

• Cohort < 200 Scans
• Big Data (Challenges and Opportunities)

Brain Research

• Brain
• Image Segmentation
• 200+ Feature variables extraction from the images

Abdomen

• Images - Segmentation - DL
• Develop Clinical Applications

Multi-Altlas Segmentation Framework

• State of art before DL
• Manually Label
• Apply deformation field
• Similar to Adaboost (Several Weak learners merge for stronger results with combinations)

Multi-Atlas Label Fusion

• Voting Label Fusion
• Majority Vote

4D Longitudinal Joint Label Fusion

• Segmentation and Results
• PCA to reduce dimensionality

Deep Machine Learning

• SLANT proposed
• Dataset Sources 

Probablistic Atlases

Multi-atlas CRUISE (MaCruise)
Analysis Brain Volume Vs Aging

Abdomen

• Image Processing
• MRI Data
• Segmentation of parts

• Segmentation is Classification Problem
• Spatial Invariance
• Localization

• GCN with larger Kernel

• GAN Application in computing

Image Synthesis

Classification and Landmark Detection

Key Learning's

• Image Segmentation
• Regression Analysis
• Cubic Spline Regression
• Deep Learning for Image Segmentation
• GAN Applications
• GCN - Global Convolutional Network - Paper

AI in Medical Imaging
Medical

• Different image sizes for X-Rays, CT-Scans
• Stacked up images, Scaling problems
• Understand in medical context
• Classification, Segmentation

Happy Mastering DL!!!

Day #180 - ML In Medicine - Data Related

Today's learning is from Phd Thesis - "Improving diagnosis of acute coronary syndromes in an emergency setting: A machine learning approach"

Key Lessons

• Arteris clogged, Blood cannot transfer oxygen
• Electrocardiogram, ECG of timeseries data
• ANN - Feed the features from ECG - Output is the probability of acute cornonary syndrome
• Ensembles of Networks
• Clinical Trials

Features Selected

• Age
• Sex
• Smoking Status
• Diabetics Status
• Cholesterol Levels
• Troponin Levels

Explanation Methods

• Input Sensitivity Analysis
• Generalized Odds Ratio
• Euclidean distance measure
• Iterative input clamping

Detailed Results Study based on Clinical Trials. After this I quickly looked up for available medical datasets.

Heart Disease Data Set 
Lung Cancer Data Set
Primary Tumor Data Set 
Acute Inflammations Data Set 
Breast Cancer Wisconsin (Original) Data Set 

My Wish - These stand-alone datasets can be used to predict considering our Full Body Medical Checkup results and map it to know the probability of any disease :)



Next Talks
The Future of Machine Learning in Clinical Imaging
Case Study: TensorFlow in Medicine - Retinal Imaging (TensorFlow Dev Summit 2017)
PhD: Machine Learning for medical Image Analysis
Deep Learning in Medical Imaging - Ben Glocker
Introduction to Deep Learning, Keras, and TensorFlow
Gated-Dilated Networks for Lung Nodule Classification in CT scans

Happy Mastering DL!!!

Day #178 - Computer Vision for Diet Assessment - Good Startup Idea

Key Lessons

• Automate Diet using Computer Vision
• Computer information -> Nutritional output

Analysis

• What is Food Item
• What is the Quantity

Image Techniques

• Recognition (Classify Food Items)
• Segmentation (Proportion of Each Item)
• 3D Model

Steps

• Identify Food
• Size and Shape to Capture proportions
• Find Nutrients

User Input - Food Images
System Functionality

• Food Image
• Detect Dish
• Segment Dish
• Recognize Food
• 3D Model
• Nutrients Content

Detection and Segmentation

• Image - Find Edges, RANSAC
• Random Sampling, Group Edge points by proximity
• Automatic Segmentation, Semi-Automatic Segmentation
• Compute Gradient Map (CNN to detect semantic Edges)
• Identify Dishes and Objects

Reconstruction Techniques

• Intensity Range Correction
• Ordinal matching filters
• Ordinal Depth range from matches
• Hierarchical reconstruction with robust matchers
• Ellipse for calibration optimisation
• Circle Calibration 
• Algebric Work
• Dense Matching using Normalized Cross Correlation

My Thoughts - OpenCV, Contour Detection

Paper to Check - Food Image Segmentation for Dietary Assessment

Volume Estimation

• Depth Scans
• Heavy Algorithm
• Method Manually Segmented
• Single View, Multiple-View, Dense Multi-View
• Food / Non-Food Separation 
• Semantic Structures
• Paper Check - Computer vision based Corbohydrate estimation type
• Clinicals trials, ground truth comparison
• Paper Check - Mobile Phone Based Estimation Vs Self Checks
• Identify Metrics of Health 

Summary - This talk provides good clarity on areas to utilize OpenCV, CNN and approach for estimating the Dietary computations.

Happy

Next List of Learnings

• Improving diagnosis of acute coronary syndromes in an emergency setting: A machine learning approach
• Optimization modelling and machine learning techniques towards smarter systems and processes
• Deep Learning for Arbitrary Code Generation: Thesis Presentation
• Understanding visual appearance on the web using large-scale crowdsourcing and deep learning.
• Matrix Factorization Methods for Training Embeddings in Selected Machine Learning Problems

Happy Mastering DL!!!