Machine learning has led to tremendous impact on healthcare - diagnosis and treatment. Employing image classification and image segmentation various diagnostics insights & solutions with automated report generation can be delivered in real-time, leading to faster and more informed decisions & streamlining costs.

1. Data Science for Healthcare:
Image Analytics
April 6, 2017

2. Page 2 © AlgoAnalytics All rights reserved
Outline
Image Analytics Work in
Healthcare
Machine Learning and
Healthcare
About AlgoAnalytics

3. Page 3 © AlgoAnalytics All rights reserved
CEO and Company Profile
Aniruddha Pant
CEO and Founder of AlgoAnalytics
PhD, Control systems, University of
California at Berkeley, USA 2001
• 20+ years in application of advanced mathematical techniques
to academic and enterprise problems.
• Experience in application of machine learning to various
business problems.
• Experience in financial markets trading; Indian as well as global
markets.
Highlights
• Experience in cross-domain application of basic scientific
process.
• Research in areas ranging from biology to financial markets to
military applications.
• Close collaboration with premier educational institutes in India,
USA & Europe.
• Active involvement in startup ecosystem in India.
Expertise
• Vice President, Capital Metrics and Risk Solutions
• Head of Analytics Competency Center, Persistent Systems
• Scientist and Group Leader, Tata Consultancy Services
Prior Experience
• Work at the intersection of mathematics and other
domains
• Harness data to provide insight and solutions to our
clients
Analytics Consultancy
• +30 data scientists with experience in mathematics
and engineering
• Team strengths include ability to deal with
structured/ unstructured data, classical ML as well as
deep learning using cutting edge methodologies
Led by Aniruddha Pant
• Develop advanced mathematical models or solutions
for a wide range of industries:
• Financial services, Retail, economics, healthcare,
BFSI, telecom, …
Expertise in Mathematics and Computer
Science
• Work closely with domain experts – either from the
clients side or our own – to effectively model the
problem to be solved
Working with Domain Specialists
About AlgoAnalytics

4. AlgoAnalytics - One Stop AI Shop
Healthcare
•Medical Image diagnostics
•Work flow optimization
•Cash flow forecasting
Financial Services
•Dormancy prediction
•Recommender system
•RM risk analysis
•News summarization
Retail
•Churn analysis
•RecSys
•Image recognition
•Generating image description
Others
•Algorithmic trading strategies
•Risk sensing – network theory
•Network failure model
•Clickstream analytics
•News/ social media analytics
Aniruddha Pant
CEO and Founder of AlgoAnalytics
• Structured data is utilized to
design our predictive analytics
solutions like churn,
recommender sys
• We use techniques like
clustering, Recurrent Neural
Networks,
Structured
data
• We use text data analytics for
designing solutions like sentiment
analysis, news summarization
and much more
• We use techniques like Natural
Language Processing (NLP),
word2vec, deep learning, TF-IDF
Text data
• Image data is used for predicting
existence of particular pathology,
image recognition and many
others
• We employ techniques like deep
learning – convolutional neural
network (CNN), artificial neural
networks (ANN) and technologies
like TensorFlow
Image
data
• We apply sound data to design
factory solutions like air leakage
detection, identification of empty
and loaded strokes from press
data, engine-compressor fault
detection
• We use techniques like deep
learning
Sound
Data

5. Page 5 © AlgoAnalytics All rights reserved
Technology:

6. Page 6 © AlgoAnalytics All rights reserved
Healthcare: Areas of Analytics Application
Patient Aid – Medical Diagnostics
Revenue Cycle Management
Resource Allocation
Kind of proposals with higher chance of
funding

7. Page 7 © AlgoAnalytics All rights reserved
Machine Learning for Healthcare: Overview
Medical diagnostics – Detecting
serious disorders or diseases through
image analytics and medico-genomic
data
Reducing Readmissions – using scoring
techniques to locate high risk customers
and preventing readmissions through
higher monitoring
Population Health Management -
Using vast amounts of past patient
data/ genomics data for diagnosing
high risk group through risk
stratification score
Cash Flow Forecasting – Forecasting
of cash flows based on claims history,
reimbursement analysis and potential
denials to forecast cash
Billing Errors– Identify opportunities
to collect missing income, including
wrongfully rejected claims by payers
or overdue money from patients.
Patient Selection - Insurance
coverage, Identifying patients
especially those who are not likely to
pay in full
Work Flow Optimization– Using
historical data for staffing to reduce
costs, Having the right clinician at
right time at right place
Efficient Use of Hospital Resources –
Prevent bottlenecks in urgent care by
analyzing patient flow during peak
times
Fraud and abuse prevention - Cost
trending-forecasting, care utilization
analysis, actuarial and financial
analysis are commonly used
applications for preventing frauds
Trend Analysis - Tracking hospital’s
market position and competitive
trends. And visualizing changes in
market dynamics and growth
patterns..
Grant problem - Predict likelihood
that a particular proposal will receive
grant using text analytics

8. Page 8 © AlgoAnalytics All rights reserved
IMAGE ANALYTICS WORK IN HEALTHCARE

9. Page 9 © AlgoAnalytics All rights reserved
AlgoAnalytics Image Analytics Expertise: Methods and Technologies
Methods TechnologiesStatistical Learning
Artificial Neural Network
Convolutional Neural Network
Transfer Learning (Deep Learning)
TensorFlow-Python for neural networks (feed-
forward and CNN)
R Programming for statistical models: using pixel
values as features and applying models such as
logistic regression, random-forest
Google’s Inception model (pre-trained TensorFlow model on Imagenet
dataset)
•In healthcare segment, classification
algorithms help predict whether a
particular pathology exists or not.
•Image classification can be used in various
solutions like
•Diabetic ophthalmology
•Brain MRI scan
•Skin disease diagnosis
Image Classification
•Segmentation is the process of dividing
image into regions with similar properties
•Medical image segmentation aims at
studying anatomical structures, locate
tumour, lesions or other abnormaliaties
•Some applications
•Retinal vein segmentation,
•Tumour segmentation
•Aneurysm detection
•Lung cancer, breast cancer,
hepatocellular carcinoma
Image segmentation
•Gives description of what is occuring on
the given set of images, with its caption,
create a predictive model which
generates relevant caption for the unseen
images.
•We use a combination of deep
convolutional neural network (CNN) and
LSTM language modeling (RNN) to
generate captions
•This can be used in automated report
generation from image analytics
Image description

10. Page 10 © AlgoAnalytics All rights reserved
Diabetic Retinopathy Brain MRI Scan
Image Classification - Healthcare

11. Page 11 © AlgoAnalytics All rights reserved
Chest Xray Diagnostics - predict whether it is normal or nodular
Input X-ray images
VGGNet-16 Extracted featureset vectors
Deep learning
Featureset
generation
Logistic regression
Classification
Nodular (class -1) Vs. Normal(class -0) X-ray
Results
● Classification Accuracy : 75%
● AUC : 81.29%
● Kappa Score : 47.3 %
● Deep Learning Techniques :
○ Maxpool-5 layer of a VGGNet-16 deep learning neural
net to extract features from the X-rays.
○ Machine Learning algorithms (Logistic regtression,
Random forest) to perform classification of the X-rays
into 2 classes - 0 as Normal and 1 as abnormal.
Dataset :
The dataset consists of Images made available by the
Department of Health and Human Services of
Montgomery County, MD, USA.
138 posterior-anterior x-rays
Out of which, 80 are normal X-rays and 58 x-rays are
abnormal with manifestations of tuberculosis.
The set covers a wide range of abnormalities
Image Classification - Healthcare

12. Page 12 © AlgoAnalytics All rights reserved
Image Segmentation - Automatic Brain Tumor Segmentation
 Dataset :
- All MRI data was provided by the 2015 MICCAI BraTS Challenge,
which consists of approximately 250 high-grade glioma cases and 50
low-grade cases.
- Each dataset contains four different MRI pulse sequences, each of
which is comprised of 155 brain slices with each pixel
representing a 1mm3 voxel, i.e., total of 620 images per patient. An example of a scan with the ground truth segmentation
with the segmentation labels
Original Image
Normalized Image
Final N4 Bias
corrected Image
Input preprocessing
 Methodology :
- Applied normalization and n4ITK bias correction on the original input images .
- Generated 43264 patches per image for the input to the 4 layered Convolutional Neural Network (CNN) model.
- Classification of tumor area based on 4 categories viz., edema, advancing tumor, non –advancing tumor and necrotic
tumor core.
Prediction Process using CNN model
Results with segmented tumor

13. Page 13 © AlgoAnalytics All rights reserved
Segmentation of blood vessel in retina Microaneurysms in color fundus image
Image Segmentation
Objective : Microaneurysms (MA) segmentation for early diabetic
retinopathy screening using fundus images
Dataset : Publicly available Retinopathy Online Challenge (ROC) dataset (total
50 fundus images) with pixel locations and radius of microaneurysms
Deep Learning Approach :
- Creating smaller patches of input images where center pixel is
either MA or Non-MA
- Predicting the probability of center pixel being MA using
Convolutional Neural Network (CNN) model
- Using trained CNN model to locate MAs in unseen fundus image

14. Page 14 © AlgoAnalytics All rights reserved
Image Segmentation - Deep Learning for Ultrasound Nerve Segmentation
Problem Statement: Accurately identifying nerve structures in ultrasound images for
effectively inserting a patient’s pain management catheter.
U-Net U-Net Trained Model
Input
Data
Ultrasound Image
Mask
Test Image
Predicted Mask
Process:
Step 1 - Train neural network on given
ultrasound nerve images and their masks
Step 2 - Use trained model to predict the
mask for new test image
Results on 500 test images: Dice Coefficient (pixelwise accuracy) = ~72%

15. Page 15 © AlgoAnalytics All rights reserved
Benefits of Image Analytics in Healthcare
Provides real-time insight to healthcare providers
during diagnosis and treatment
Improves the standard of care and speed of diagnosis,
treatment and recovery
Reduces varying subjective interpretation and human
error
Offers the potential to make faster and more
informed decisions, streamline costs and broadly
improve the quality and economics of healthcare
Prediction of certain genetic disorders, monitor
patients at risk
Image Analytics in healthcare current estimated US market size is at $141 million and expected to grow at 8% CAGR
The Benefits include:

16. Interested in knowing more?
Contact us: info@algoanalytics.com
April 6, 2017