AI in Healthcare

Dr. Sema Candemir

Eskişehir Technical University Computer Engineering Department
Learning Objectives

Students will
• acquire both theoretical and practical knowledge in the field of
medical data analysis.
• develop solid understanding of image analysis and machine learning
in healthcare problems.
• gain hands-on experience in applying AI/ML, image processing, and
pattern analysis techniques to healthcare problem through home
works.

AI in Healthcare Dr. Sema Candemir

 Prerequisites
• Proficiency in Python, or ability to fast self-learn. Python will be used for
homework assignments and the course project.
• CME 193 - Scientific Python (stanford.edu)

• Fundamental knowledge in probability, statistics and algorithm analysis.

• Familiarity with image processing, pattern recognition and machine learning.

AI in Healthcare
 Grading
• Class attendance is mandatory 5%
• Homeworks 25%
• Midterm 30%
• Final 40%

• Grading will be straight scale (and no bell-curve).

• 90-100 AA, 85-89 AB, 80-84 BA, 75-74 BB, 70-74 BC, 65-69 CB, 60-64 CC, 55-59 CD, 50-54 DC,
45-49 DD, less than 45 FF.

AI in Healthcare Dr. Sema Candemir

Suggested Resources
• O’Reilly - Hands-On Machine Learning with Scikit-Learn, Keras,
and TensorFlow: Concepts, Tools, and Techniques to Build
Intelligent Systems https://github.com/ageron/handson-ml2

• Stanford University - Artificial Intelligence in Healthcare

https://web.stanford.edu/class/biods220/)
• https://www.coursera.org/learn/fundamental-machine-
learning-healthcare (Stanford University)

• E. Alpaydin. Introduction to machine learning. MIT press, 2020.

(mention recent YouTube broadcast)
• C. M. Bishop. Pattern recognition and machine learning.
• Machine Learning Mastery

AI in Healthcare
• Sema Candemir – [email protected]
• Office hours: Tuesdays 10-11 am Thursdays 10-11 am.

• Undergraduate Teaching Assistant

• Umut Kaan Kavaklı [email protected]
• Office hours: Mondays 14-15pm Wednesdays 14-15 pm

AI in Healthcare Dr. Sema Candemir

Digital Transformation
• With the current technology, computers can store and
process massive amounts of data at high-speed.
• (AI-based /data-driven systems / state of the art).

Revolution
• The governments, research institutes, companies have been
working on strategies, road-maps to develop and utilize AI-
based technology.
• National-Artificial-Intelligence-Research-and-Development-Strategic-Plan-2023-
Update.pdf (whitehouse.gov)
• Digital Transformation Office of the Presidency of Turkey - National Artificial
Intelligence Strategy 2021-2025 (cbddo.gov.tr)

• There is a massive investment in research & development for

AI-based technology for all sectors, including healthcare.

AI in Healthcare Dr. Sema CANDEMIR

 Digital Transformation in Healthcare
• Artificial Intelligence in Healthcare refers to the development and
diverse utilization of AI in medical and health services.
• AI technology will play a key role in the diagnosis and treatment
process of diseases, aiming to make healthcare services more effective
and efficient.

• Yapay zeka tabanlı algoritmalar, büyük miktarda tıbbı veriyi analiz

ederek, verilerden, klinik açıdan anlamlı olabilecek bilgileri ortaya
çıkarabilir.
• Bu bilgiler, klinik karar verme süreçlerinde doktorlara destek
sağlayabilir, sağlık problemlerinin çözümünde etkili olabilir ve
hastalıkların teşhis ve tedavi süreçlerinde yardımcı bir rol üstlenebilir.
AI in Healthcare Dr. Sema Candemir
 Digital Transformation in Healthcare
• Machine Learning is the fundamental part of this digital
transformation.
• Machine Learning –based algorithms analyze medical data and
transform it into clinical relevance / clues about the patient`s health.
• Machine Learning –based algorithms analyze medical data.
• Medical data contain a wealth of information that may help understand patient
health.

AI in Healthcare Dr. Sema Candemir

 Healthcare Data
• Imaging (radiology, pathology, ophthalmology, neuroradiology)
• Clinical reports
• Time series (electrocardiograms)
• Lab results

• Structured clinical data: have a consistent organization like a table with rows and columns
• Unstructured clinical data: images, clinical notes, electrocardiograms, speech.

AI in Healthcare Dr. Sema Candemir

Machine Learning –based algorithms analyze medical data and transform it into clinical relevance / clues about the
patient`s health.

• CheXNet: Radiologist-Level Pneumonia Detection on Chest

X-Rays with Deep Learning (stanfordmlgroup.github.io)

AI in Healthcare Dr. Sema Candemir

• Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in
Retinal Fundus Photographs | Diabetic Retinopathy | JAMA | JAMA Network
• Dataset: 28175 retinal images
• Graded by a panel of 54 US licensed ophthalmologists
• The algorithm computes diabetic retinopathy severity from the intensities of the pixels in the retinal
images.

• Frontiers | An overview of artificial intelligence in diabetic retinopathy and other ocular diseases
(frontiersin.org)
• Classification-based screening of Parkinson’s disease patients through voice signal | IEEE Conference Publication
| IEEE Xplore
• Classification-based screening of Parkinson's disease patients through voice signal - YouTube

AI in Healthcare Dr. Sema Candemir

 Automated Cough Sound Analysis for Detecting Childhood
Pneumonia | IEEE Journals & Magazine | IEEE Xplore

AI in Healthcare Dr. Sema CANDEMIR

 Deep learning and the electrocardiogram: review of the
current state-of-the-art

AI in Healthcare Dr. Sema Candemir

 Medical Data Analysis
Automating tasks that clinicians can do by analyzing medical data

Medical image analysis is an interdisciplinary domain which integrates knowledge in image

processing, machine learning, computer vision, statistics, mathematics, imaging informatics
and healthcare.

• With the current technology, computers can store and

process massive amounts of data at high-speed.
• (AI-based /data-driven systems / state of the art).

Artificial Intelligence/Machine Learning has demonstrated

tremendous advancements in recent years in healthcare.

AI in Healthcare Dr. Sema Candemir

 • Dataset: 28175 retinal images
• Graded by a panel of 54 US licensed ophthalmologists
• The algorithm computes diabetic retinopathy severity from the
intensities of the pixels in a fundus image.

Both studies reported Clinician-level performance !!

AI in Healthcare Dr. Sema Candemir

• 3D Convolutional Neural Network
• Analyzing coronary computed
tomography angiography for
automated detection of
atherosclerotic plaques in vessels.
• Discriminatory appearance features
between vessels with and without
atherosclerotic plaque.
• Detecting and Characterizing Inferior Vena Cava Filters on
Abdominal Computed Tomography with Data-Driven
Computational Frameworks | Journal of Imaging Informatics in
Medicine
• Currently, determining the location and the type of IVC filters
are done by interventional radiologist.
• We propose using machine learning and image processing
algorithms to analyze abdominal CT scans for automated
detection of IVC filters and determination of the type of filter.
• The developed model might turn into a diagnostic decision
support tool for interventional radiology, which may reduce
the reading time and help filter management decision.

Dr. Sema CANDEMIR

 • Review and evaluate the ~100 AI/ML studies in
brain MRI analysis over the past 12 years
(2022)
• 71% - classification of AD, MCI and (HC-health
control) normal aging
• 25% - prediction of MCI conversion to AD

• Support Vector Machines (SVM): a conventional ML

model, determine optimal decision boundaries for
classifying data.
• Random Forest (RF): combination of decision trees.
Robust. Better on tabular data.(structured
data)….Easier to program but computationally
costly.
The best performance was achieved by using deep • Logistic Regression (LR): generated probabilities for
learning-based convolutional neural network (CNN) with binary dependent variable.
weighted accuracy 89%. • CNN: Convolutional Neural Networks
HC vs. AD classification: the reported performance, the • NC-ANN: Non-convolutional artificial neural
weighted average accuracy is 91% networks
MCI to AD conversion: the weighted average accuracy of
79%. AI in Healthcare Dr. Sema CANDEMIR
• 2022 August – IEEE J Biomedical Health Informatics
• AD vs. HS – find the distinct anatomical landmarks
• Extract patches to fed into CNNs
• Fine-tune model with pMCI and sMCI

Experimental results on the ADNI-1

dataset indicate that proposed
method outperforms existing
methods for MCI classification, with
an 79.9% accuracy and 85.96% F1-
score.

3rd International Balkan Congress on Neurology and Neuroscience Dr. Sema CANDEMIR
 Retina as a window to the brain

• There are also studies which detect cognitive diseases from

the retinal images.
• Cheung, C Y., et al. "A deep learning model for detection of
Alzheimer's disease based on retinal photographs: a
retrospective, multicentre case-control study." The Lancet
Digital Health- 2022
• Hypothesis: AD related features on retinal images are
present on either one eye or both eyes.
• Deep Learning algorithm which use retinal photograph
alone to detect AD.
• 648 patients with AD (5598 images)
• 3240 people without AD (7351 images)
• From 11 clinical center from different countries (Hong
Kong, China, Singapore, and the USA)
3rd International Balkan Congress on Neurology and Neuroscience Dr. Sema CANDEMIR
• A bilateral deep learning model, which learns AD-related features from
optic nerve head-centered and macula-centered retinal images from
both eyes.
• Efficient Net as the backbone feature extractor
• Tan M, and Le, Q. Efficient Net: Rethinking Model Scaling for Convolutional Neural
Networks. Proceedings of the 36th International Conference on Machine Learning
2019.

the bilateral deep learning model had

accuracies ranging from 79·6% to 92·1%

3rd International Balkan Congress on Neurology and Neuroscience Dr. Sema CANDEMIR
• Proof-of-concept study
• Researchers at Duke University Medical Center
• June 25, 2023 – Ophthalmology Science (Elsevier)

• Machine learning model that can differentiate

normal cognition from mild cognitive impairment
using optical coherence tomography angiography
(OCTA) along with patient data such as age, sex,
visual acuity, and years of education.
• The model analyzes retinal images (and
associated data) and recognize digital biomarkers
to identify individuals with MCI.

3rd International Balkan Congress on Neurology and Neuroscience Dr. Sema CANDEMIR
A Report to the Board of Scientific Counselors April 2015:
Automated Detection of Lung Diseases in Chest X-Rays (nih.gov)

AI in Healthcare Dr. Sema Candemir

 Digital Transformation in Healthcare
• Artificial Intelligence in Healthcare refers to the development and
diverse utilization of AI in medical and health services.
• AI technology will play a key role in the diagnosis and treatment
process of diseases, aiming to make healthcare services more effective
and efficient.

• AI-based/ML–based algorithms analyze medical data and transform it

into clinical relevance / substantial findings about the patient`s health.

AI in Healthcare Dr. Sema Candemir

 Digital Transformation in Healthcare
• Research Institutes and Medical Centers
• The U.S. National Institutes of Health
• The OSU Wexner Medical Center, Massachusetts General Hospital, Children`s National Hospitals….
• Companies
• Siemens Healthineers, Nvidia Healthcare, IBM Health, …(data scientist + research scientist)
• Universities

• TUSEB - Health Institutes of Turkey

• Saglik Veri Araştırmaları ve Yapay Zeka Uygulamaları Enstitüsü
• https://www.resmigazete.gov.tr/eskiler/2022/03/20220312-9.htm
• https://tuyze.tuseb.gov.tr/
• Tubitak
• Universities
• https://aihealthlab.eskisehir.edu.tr/

AI in Healthcare Dr. Sema Candemir

 Computational Framework
Regions of Interest Detection
organs, nodules, IVC filters, pathology
lesions
Clinical
Problem Computational Disease Detection
Framework pneumonia , TB, Covid-19
Medical Data
• Images (CXR, CT, Ultrasound,…) Prognosis
• Clinical Reports (Radiology reports) Severity, mortality, cognitive decline, fetus
• Medical Health Records development,…
• Demographics

Computational Framework (mathematical operations)

Image processing, Computer vision, Machine Learning, Statistics, Programming skills (C++, Python), Libraries (tensor
flow, keras…), follow up recent literature…
AI in Healthcare Dr. Sema Candemir
What is AI?
• Intelligence: The capacity for logic, understanding, self-awareness, emotional knowledge, reasoning,
planning, creativity (e.g. Dall-E), critical thinking, and problem solving.
• Artificial Intelligence: Broad term, making machines to imitate human intelligence and intelligent
behaviors
• "AI as the science and engineering of making intelligent machines, especially intelligent computer
programs" - John McCarthy – Stanford University

Artificial Intelligence
Machine Learning (Linear
regression, SVM, Decision Trees…)

Deep Learning
(CNN, GANs, RNN…)

AI in Healthcare Dr. Sema Candemir

• Machine Learning: developing a model/algorithm
which learns from data (without being explicitly Artificial Intelligence
programmed ) and make inference for unseen
instances. (Support Vector Machines, Linear Machine Learning (Linear
Regression, Decision Trees…) regression, SVM, Decision Trees…)

Deep Learning
• Neural Network: a type of ML algorithm, which (CNN, GANs, RNN…)
consists of nodes each one process the data and passes
the output to the next node. NN are organized in
layers.

• Deep Learning: subset of machine learning/uses neural

networks (architectures with many layers) for the
learning process.

A biological neuron receives many inputs and

fires a signal if sum of the inputs high enough.
AI in Healthcare Dr. Sema CANDEMIR
 Other Related Disciplines: Pattern Recognition – Pattern Analysis

• The process of identifying patterns in data.

• A pattern is a type of theme of recurring events or objects. Depending on the
application, these recurring information can be in images, signal waveforms, or
text.
• Each person`s face is a pattern which is composed of a particular combination of eyes, nose,
mouth, eyebrows.

• Speaking

• Handwritten characters

• Patterns in Diseases

Dr. Sema Candemir

 Other Related Disciplines: Computer Vision
• Mimic human visual understanding and make computers as capable as humans for perceiving the 3D
dimension.
• Developing mathematical techniques (theoretical and algorithmic basis) for extracting the 3D information from
Images (2D) .
• From the engineering point of view, computer vision aims to build autonomous systems to perform some of
the tasks which the human visual system can perform and even surpass it in many cases.

Computer Vision: Algorithms and Applications, 2nd ed. (szeliski.org)

• Computer Vision: Algorithms and Applications, 2nd ed. (szeliski.org)

• Shapiro, Linda G., and George C. Stockman. Computer vision. Pearson, 2001.
• Introduction_t0_ComputerVision_at_GeorgiaTech
• LearnOpenCV – Learn OpenCV, PyTorch, Keras, Tensorflow with examples and tutorials
• UCF CRCV - YouTube
AI in Healthcare Dr. Sema Candemir
 • Where is the vena cava filter in the Computed
Tomography sequences?

• Where is the tumor in a

mammogram?

AI in Healthcare Dr. Sema Candemir

Other Related Disciplines: Image Processing
• Image processing is the use of algorithms to perform
operations on digital images.
• Image enhancement, restoration, image compression,
image segmentation, rescaling image, correcting
illumination, etc.…
• Image In – Image Out
• The goal is to improve the quality of an image, extract Image Credit: https://ai.stanford.edu/~syyeung
useful information from it.

Image Credit: mathworks.com

AI in Healthcare
• Gray scale image is a matrix of pixels (picture elements).
• Each pixel stores its intensity ranging.
• Images are composed a group of pixels (in 2D) or a group of voxels (in 3D)
• Most medical imaging types like X-rays, CT scans, MRI, and ultrasound are all performed in grayscale.
• We also see color images in many areas of medicine like dermatology, ophthalmology, and pathology
for example.

AI in Healthcare Dr. Sema Candemir

 Computational Framework
Regions of Interest Detection
organs, nodules, IVC filters, pathology
lesions
Clinical
Problem Computational Disease Detection
Framework pneumonia , TB, Covid-19
Medical Data
• Images (CXR, CT, Ultrasound,…) Prognosis
• Clinical Reports (Radiology reports) Severity, mortality, cognitive decline, fetus
• Medical Health Records development,…
• Demographics

Computational Framework (mathematical operations)

Image processing, Computer vision, Machine Learning, Statistics, Programming skills (C++, Python), Libraries (tensor
flow, keras…), follow up recent literature…
AI in Healthcare Dr. Sema Candemir
 How do Machines Make a Decision?
• Problem Definition: Casting the clinical
problem as a ML algorithm
• Curate Data Clinical Data Curation Data
Problem Model Training
• Obtain annotations (e.g., reference Definition
(e.g., Brain
MRI images)
Annotation
standard, ground truth).
• Model Training Strategy (e.g., supervised, AI-based Prediction
semi-supervised, unsupervised… ) Algorithm
Model Evaluation
• Process data
• Evaluation/ Test on External dataset
• Deployment/Clinical effectiveness in real- Prediction
regarding the
world settings patient’s health
Medical Data
obtained from a new AI-based Prediction
patient Algorithm

AI in Healthcare Dr. Sema Candemir

 Course Syllabus (tentative)
• Lecture1 - Introduction to AI in Healthcare
• Lecture2 - Setting Up Development Environment
• Lecture3 - Introduction to Machine Learning for Medical Data Analysis
• Lecture4 - Computational Framework
• Clinical task – Data curation--- Data annotation--- Model training – Model Evaluation

• Lecture 5 - Model Training Loop

• Training a model –Loss functions – Optimizers (Gradient Descent) - Learning rate.
• Lecture 9 - CNN
• Lecture 6 - Introduction to Deep Learning
• Lecture 10 – Model Training Strategies
• Lecture 7 - Medical Data • Supervised – Semi-Supervised – Unsupervised – Transfer Learning – Weak supervision
• Lecture 8 - Preparing Data for Processing • Lecture 11 – Model Evaluation
• Lecture 12 – Dealing with Limited Medical Data
• Overfitting – Underfitting – Generalization – Learning Curves – Early Stopping
• Mitigating Overfitting – Data Augmentation – Dropout – Batch normalization

• Lecture 13 – Medical Image Applications: Segmentation

• Lecture 14 - Medical Image Applications: Classification and Detection
• Lecture 15 - Medical Image Applications: Continuous Data
• Lecture 16 – Explainable AI
• Lecture 17 – Mobile Health (weight pruning techniques)