The document discusses federated learning as a decentralized machine learning method beneficial for medical imaging, focusing on its ability to preserve data privacy while allowing device collaboration without data sharing. It outlines the benefits, limitations, and challenges of federated learning, including issues related to data heterogeneity and communication overhead. The document also reviews various studies and algorithms related to federated learning, emphasizing its potential applications in healthcare and future directions for advancement.

1. FEDERATED LEARNING
FOR
MEDICAL IMAGING
SUBMITTED BY :
Group No. 10
Guide Name – Dr Rashmi Ranjan Sahoo
Ankita Patra(2101109131)
Ansuman Simanta Shekar Bhujabala(2101109132)
Hritik Dash(2101109143)
Purnima Prusty(2101109185)

2. Introduction to
Federated
Learning
Federated learning is a machine learning technique that trains
models on decentralized data. It allows devices to collaborate
without sharing their data.

3. LIMITATIONS OF USING DEEP LEARNING APPROACHES IN
MEDICAL IMAGING
Data Privacy
Concerns:
• Sensitive
Information
• Regulatory
Compliance
Data
Centralization
Issues:
• Data Security Risks
• Infrastructure
Costs
Limited Data
Access:
• Data Silos
• Collaboration
Barriers
Bias and
Generalization:
• Limited Diversity
• Sample Size
Limitations
High
Computational
Costs:
• Resource Intensive
• Scalability Issues
Model Updates
and
Maintenance:
• Frequent Updates
Needed
• Coordination
Challenges

4. BENEFITS OF USING FEDERATED LEARNING FOR MEDICAL IMAGING
Data Privacy
Devices don't need to share their raw data with each other or a central server.
Data Security
It reduces the risk of data breaches or unauthorized access to sensitive information.
Scalability
It can effectively train models on massive datasets distributed across numerous devices.
Flexibility
It allows models to be trained on data from different sources and domains.

5. LITERATURE SURVEY
SL NO. TITLE YEAR OF
PUBLICATION
JOURNEL AUTHOR TOPIC
1
A Systematic Review on
Federated Learning in
Medical Image Analysis
2023 IEEE Access
Md
Fahimuzzman
Sohan And
Anas
Basalamah
The systematic review by M. F.
Sohan and A. Basalamah explores
the application of Federated
Learning (FL) in medical image
analysis, emphasizing its ability to
preserve patient privacy through
decentralized data handling. The
review follows PRISMA guidelines
to assess various studies,
highlighting the performance of FL
frameworks, challenges such as
data heterogeneity, and the need
for improved privacy measures. It
concludes with recommendations
for future research to enhance the
effectiveness and implementation
of FL in healthcare settings.

6. SL NO. TITLE YEAR OF
PUBLICATION
JOURNEL AUTHOR TOPIC
2
Federated Learning in
Medical Imaging:
Part I: Toward
Multicentral Health
Care Ecosystems
Federated Learning in
Medical Imaging:
Part II: Methods,
Challenges, and
Considerations
2022
Journal of
the
american
college
of
radiology.
Darzidehkalani E,
Ghasemi-Rad M,
Van Ooijen PM
The book explores how federated learning
(FL) can address the challenge of limited
data for training deep learning models in
medical institutions by enabling
collaboration without sharing sensitive
patient data. Part I focuses on FL's role in
creating multicentral healthcare
ecosystems, enhancing patient care, and
promoting global cooperation. Part II
covers the technical aspects, applications
in radiology, and future developments of
FL. Together, the book highlights FL's
potential to improve medical imaging
through collaborative efforts while
ensuring data privacy and security.

7. SL NO. TITLE YEAR OF
PUBLICATION
JOURNEL AUTHOR TOPIC
3
Federated learning
and differential
privacy for medical
image analysis.
2022 Scientific
reports
Adnan M, Kalra
S, Cresswell JC,
Taylor GW,
Tizhoosh HR
The document explores the use of
differentially private federated learning for
medical image analysis, focusing on
histopathology images. It demonstrates
the framework's feasibility in analyzing
medical images while preserving patient
data privacy and addresses the impact of
data distribution and the number of
healthcare providers on the learning
framework's performance. The study
emphasizes the importance of differential
privacy in providing quantitative bounds
on privacy in collaborative learning.

8. • Challenges in Federated Learning
• Data Heterogeneity: Clients might have data with
different distributions, affecting model
performance.
• Communication Overhead: The frequent exchange
of model updates can be resource-intensive.
• Security Risks
• Federated Averaging (FedAvg)
• Single Weight Transfer (SWT)
• Cyclic Weight Transfer (CWT)
• Ensemble Methods
• Advantages: Preserves data privacy, leverages
diverse datasets, and can improve model
robustness.
• Disadvantages: Can suffer from performance issues
due to data heterogeneity, potential biases, and
higher communication load.
• Key points include
• Data Privacy: FL preserves patient data privacy by
keeping data locally at each institution and only
sharing model updates.
• Collaboration: It facilitates multi-institutional
collaborations without needing a centralized
database, overcoming legal and logistical barriers.

9. FEDERATED LEARNING ALGORITHMS
Federated Averaging
This is a basic algorithm where
devices compute model
updates locally and then send
them to a central server to be
averaged.
Federated Stochastic
Gradient Descent
This algorithm uses stochastic
gradient descent to update
models iteratively, but with
decentralized data. It's similar
to Federated Averaging, but
with additional features for
efficiency and performance.
Federated Proximal
This algorithm is used for
constrained optimization
problems, ensuring that the
model updates remain within a
defined constraint.

10. FEDERATED AVERAGING
1 Local Training
Each device trains a model on its own data.
2 Model Aggregation
A central server collects the model updates
from all devices.
3 Global Model Update
The server averages the model updates and
distributes the new model to the devices.

11. FEDERATED STOCHASTIC
GRADIENT DESCENT
1 Stochastic Updates
Each device uses stochastic gradient descent to update the model, using only a
subset of its data.
2 Noise Reduction
The algorithm introduces techniques like momentum and adaptive learning
rates to reduce the noise in the stochastic updates.
3 Communication Efficiency
It optimizes communication by sending smaller updates to the server.
4 Convergence
The algorithm aims to converge to a globally optimal model, even with
decentralized data.

12. FEDERATED PROXIMAL
Problem Setup
The problem involves minimizing a loss function while satisfying certain
constraints.
Local Updates
Each device computes a local update that minimizes the loss function, subject
to its local constraints.
Server Aggregation
The server aggregates the local updates and applies a projection operator to
ensure that the global model satisfies the overall constraints.
Model Distribution
The server distributes the updated model to all devices for further local
training.

13. APPLICATIONS OF FEDERATED LEARNING
Healthcare Train models for disease prediction, personalized
medicine, and medical imaging.
Mobile Devices Improve device performance, personalize user
experiences, and detect fraud.
Finance Develop fraud detection systems, risk assessment
models, and personalized financial advice.
Internet of Things Optimize resource allocation, predict equipment
failures, and enhance user experiences.

14. FEDERATED LEARNING FRAMEWORKS
TensorFlow Federated
A framework designed for federated learning,
providing tools for model development, training, and
evaluation.
Flower
An open-source framework for decentralized machine
learning, with a focus on flexibility and customization.

15. TENSORFLOW FEDERATED
Codebase
Provides a high-level API for creating and training federated learning models.
Scalability
Handles large-scale federated learning scenarios with numerous devices and datasets.
Visualization
Provides tools for visualizing the training process and model performance.
Analysis
Offers mechanisms for analyzing the training data and evaluating the model's performance.

16. FLOWER
Flexibility
Supports various federated learning strategies and custom algorithms.
Customization
Allows for modifications and extensions to meet specific needs.
Open Source
It's an open-source framework, allowing for community contributions and collaboration.
Documentation
Provides comprehensive documentation and tutorials to guide users.

17. Conclusion and Future Directions
Federated learning is a transformative technology with the potential to revolutionize
machine learning while addressing data privacy concerns. As research and development
continue, we can expect to see even more innovative applications and advancements in
this exciting field.
CONCLUSION

18. FUTURE DIRECTIONS
1. Advancements in Federated Learning
o Improved algorithms for better model aggregation
o Enhanced security measures (e.g., differential privacy)
2. Integration with Other Technologies
o Combining with edge computing and blockchain
3. Broader Adoption
o Wider use in various medical fields