Distributed Learning in Healthcare: Application of Federated

Learning to Skin Cancer Diagnosis

Yeke Zhang

Automation, Northeastern University, Shenyang, China

Keywords: Federated Learning, Skin Cancer Diagnosis, Health System.

Abstract: Skin cancer is one of the deadliest diseases, but some of its types can be treated and cured if diagnosed in

early stages. Machine learning (ML) is important for accurate skin cancer detection. Using ML to train a

module is definitely helpful to image recognition and cancer prediction. However, patients’ relevant data is

private and sensitive. It is illegal for module trainers to transport raw data directly. To solve this problem,

federated learning (FL) provides a grand new approach to construct an accurate but private skin cancer

detection system. This paper introduces the theory how FL is applied to skin cancer diagnosis and reviews

the development of FL and skin cancer detection in recent years. The paper mainly focuses on certain

outstanding applications, especially those have been proven effective and better than traditional method.

Besides, through discussion on limitations and challenges of FL in this field, the paper explores the future

direction of research. The aim is to highlight the potential of FL in the skin cancer diagnosis and application

to future health system.

1 INTRODUCTION

Machine Learning (ML) has become a very important

method in many fields over the past decade (Liu,

2021; Liu, 2023; Qiu, 2020; Qiu, 2022). Based on

large-scale training databases, machine learning

algorithms learn from the data and are able to classify,

make prediction and so on. With the rapid

development of ML, a lot of neural networks are

proposed such as AlexNet (Yuan, 2016), VGG

(Tammina, 2019), GoogLeNet (Yu, 2022). They

perform well in vision tasks. And other ML algorithm

are also used into production. For example, Support

Vector Machine (SVM) in rice prediction (Liakos,

2018). Machine learning is applied into various

realms and seen outstanding result.

Medical image classification plays a significant

role in cancer detection and risk prediction. In recent

years, machine learning has been applied into this

field and make a good contribution. It improves the

process and help doctors to recognize whether it is

cancer through learning past cancer images and data.

By this way, ML increases speed and accuracy of

diagnosis. For instance, melanoma is an extremely

deadly skin cancer that may cause 10 million fatalities

https://orcid.org/0009-0006-4885-6889

in 2020 according to global statistics. However, due

to combination of ML and medicine, melanoma

diagnostic accuracy increased from 50% to 75%

(Yaqoob, 2023).

However, traditional ML algorithms have a

common feature, they all update and transform

through a central node, which is usually the

administrator’s computer. It is available to finish the

machine learning task, but it may undermine the

patient's right to privacy. It also may create a data

breach (Hameed, 2021). The medical data contain

patients’ biological information and family privacy,

their leak may cause harmful influence to patients. In

the decades, privacy laws are getting stricter, ML

algorithm that is applied into the medical field should

consider more about privacy and traditional ML

algorithm is not enough in this regard.

In this case, federated learning can be considered

since it mainly focuses on solving privacy problem. It

provides a platform that allows clients to train model

on their own computer and only give a completed

model about their own data. During this way, users’

information avoids leaking and the central

administrator can collect all necessary information to

train and update model. Using federated learning

Zhang, Y.

Distributed Learning in Healthcare: Application of Federated Learning to Skin Cancer Diagnosis.

DOI: 10.5220/0012967700004508

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence (EMITI 2024), pages 675-678

ISBN: 978-989-758-713-9

675

algorithm, model trainer can train model based on

limited information so that they need not to worry

about privacy law because the raw information is not

sent (Chowdhury, 2021).

Many researches have been done to develop such

methods. For example, Xu et al. adopted federated

learning to achieve fairness in dermatological disease

diagnosis (Xu, 2022). Wicaksana et al. combined

federated learning with image classification to better

work (Jeffry, 2022). What’s more, one research

develops a possibility to cooperate between different

hospitals through federated learning (Hosseini, 10).

The author will focus on follow points: 1) The

development of cancer prediction application of

federated learning. 2) The theory how federated

learning are adopted into cancer detection. 3) The

limitation and challenge of this technical.

2 METHOD

2.1 Introduction of Federated Learning

Federated learning is a process that is mainly based

on the distributed learning. At first, the administrator

distributes the old model to the nodes, which are the

independent clients. Then, each client uses this model

to process the data and calculates the bias, usually

gradient. By neural network, the algorithm can

aggregate the weights and bias. The next step is

returning the processed data instead of clients’ raw

data. Finally, the administrator receives the processed

data and update the model. It is a round of machine

learning. After many rounds, the model will be more

and more accurate and practical.

Federated learning also has many optimizations.

Based on these optimization methods, optimized

federated learning performs better on a certain target.

For example, some researchers use offline

calculation, which changes modulo power operations

to modulo multiplication operations and speed up

computing. If module constructors focus on sparse

data, they can adopt sparse matrix calculation and

sparse bar chart optimization. During data

transportation, many encrypt algorithms can be

applied to protect the data security or reduce the scale

of transported data. With the development of

federated learning, more and more new methods are

invented to optimize federated learning to adapt to

certain task.

2.2 Federated Learning-Based Skin

Cancer Detection

Federated learning has demonstrated its significant

impact in the field of skin cancer diagnostics through

numerous successful applications. By integrating

with a variety of techniques, federated learning

enhances the diagnostic process for skin cancer.

Haggenm et al. develop a method for melanoma

that is a kind of serious skin cancer diagnostics by

decentralized federated learning. They pretrained

models on ImageNet. Then they used the tree-

structured Parzen estimator30 to choose the

hyperparameters to maximize the area under the

receiver operating characteristic curve (AUROC).

During the process, they increased the learning rate at

the former part and decreased the learning rate at the

latter part. And the federated data come from five

independent hospitals. All data are protected. Finally,

researchers found that decentralized method

performed better than traditional federated learning

method. It had lower AUROC on the test dataset.

Compared with traditional algorithm, it is a good

alternative (Hekler, 2024).

Ain et al. also succeeded to develop a method to

predict skin cancer. They combined many private

hospitals to support their project and got data. Every

private hospital is a client, so they train their own

patients’ data on their own computer using Support

Vector Machine (SVM) and Convolutional Neural

Networks (CNN). And they use certain algorithm to

transform the data into weight that can be

communicated among different hospitals. At the same

time, nodes transport their data to the central server.

At last, central sever can update the model to predict

the skin cancer. The system also uses median filtering

method, watershed algorithm, gray Level co-

occurrence matrix (GLCM) feature technique and

ABCD rule. Researchers used this kind of method to

test dataset and got a high accuracy. It performs a

certain level of robustness and fitness. By now, it is

one of the most effective methods as a classifier or

predictor (Ain, 2023).

Based on federated learning, a new method is

developed. An asynchronous and weighted approach

is used to help federated learning for skin lesion

diagnosis. Yaqoob et al. adopted FedOpt approach to

reduce communication overhead. They creatively

develop an asynchronous technique to update shallow

and deep lays at different rates. Using this method,

doctors can better find skin diseases. What’s more,

communication costs are reduced so it is really

economical.

FedPerl is another good example that federated

learning is used to improve skin classification. This

method improves its model through anonymous

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

676

peers. By this kind of data transportation, patients’

data will be safe. Bdair et al. firstly built

communications, then focused on peer learning and

peer anonymization. Finally, they got a good result.

The test shows that their model has a better parameter

and lower communication cost. This means a grand

new way to classify skin disease and diagnose the

most serious cancers (Bdair 2021).

Hashmani et al. developed a method to apply

federated learning into intelligent dermoscopy

device. Traditional algorithms are usually based on

visual pattern recognition of morphological features,

so this new method provides a general solution to

different images. Researchers use federated learning

to support intelligent dermoscopy device and attain

better performances and universal application

(Hashmani, 2021). With the help of this technique,

doctors can better recognize different skin legion,

which will definitely contribute to the skin cancer

detection because skin cancer usually signals beneath

the surface of skin.

3 DISCUSSIONS

Federated learning has achieved a great deal in the

skin cancer field, it completes the task to train a

practical model without transforming raw data

directly and protects the clients’ privacy. However,

there are still some limitations, especially

interpretability and poor performance compared to

traditional visual methods.

Interpretability: Due to privacy laws, different

clients share different types of data. It means that

individual clients can benefit from the collaborative

training only if their data is compatible with that of

other participating institutions (Roschewitz, 2021).

For instance, different collected different patients’

data because of privacy laws, but not all the data is

suitable to be used to train model. Besides, federated

learning may cause client data is not interoperable.

The interpretability makes federated learning face

many challenges in front of the skin cancer treatment

and diagnosis. Although iFedAvg improved the

phenomenon to a large extent, it still haves non-

negligible false-positive rate.

Poor performance compared to ABCDE rule:

ABCDE rule (Duarte, 2021) is a traditional rule to

distinguish between benign and malignant lesions.

ABCDE rule examines skin cancer according to

color, lesion scale and so on. It has been proven trusty

and fundamental. Although method based on

federated learning is effective, it still shows more

prediction mistakes than traditional ABCDE rule. The

reason mainly is pattern recognition complexity for

malignant lesion characteristics in medical imaging

(Riaz, 2023). Because of this, the method is still

thought of as a black-box method. In practice, doctors

prefer to choose ABCDE rule to visually examine the

skin cancer.

To optimize federated learning in skin cancer

detection and treatment, follow aspects may be

considered: increasing data balance, improving data

interpretability and optimizing strategies in different

stages of the disease. Above aspects directly linked to

the application of the algorithm, breakthroughs in

these areas would contribute significantly to

development of combination of federated learning

and skin cancer.

Although federated learning has many challenges,

it has potential for researchers to explore. There is

still a lot of room for optimization between

algorithms and practical applications. In addition, the

hardware situation and transmission mechanisms

should be also improved to combine with federated

learning algorithms well (Deng, 2019; Deng, 2023;

Sugaya, 2019). Much research is ongoing to improve

federated learning in learning rate, privacy protection,

data heterogeneity and resource cost. Many more

algorithms should be combined with federated

learning to solve practical problems. For example,

FedDecorr are created to solve dimensional collapse

in the learning process (Shi, 2023). FedPerl was

created to make skin cancer lesion classification more

accurately. In short, further investigations are

required to develop federated learning and skin

cancer.

4 CONCLUSIONS

This paper overviews recent developments of

federated learning for skin cancer, and discusses the

applications of federated learning in kinds of skin

cancer treatments or diagnosis. Federated learning

has a successful application, which offers a

considerable method to improve skin cancer

treatment system and doesn’t against the privacy

laws. Many algorithms and machines use federated

learning to improve their performance and protect

clients’ privacy. However, using federated learning to

improve skin cancer prediction or treatment will also

have some challenges such as more time spending

and poor interpretability. Besides, prediction

accuracy and model complexity also need to improve.

So, aiming to have an adequate understating of

application of federated learning for skin cancer,

further research and investigation are needed to find

out potential benefits and challenges.

Distributed Learning in Healthcare: Application of Federated Learning to Skin Cancer Diagnosis

677

REFERENCES

Ain, Q. U., Khan, M. A., Yaqoob, M. M., Khattak, U. F.,

Sajid, Z., Khan, M. I., & Al-Rasheed, A. 2023. Privacy-

Aware Collaborative Learning for Skin Cancer

Prediction. Diagnostics, 13(13), 2264.

Bdair, T., Navab, N., & Albarqouni, S. 2021. FedPerl: semi-

supervised peer learning for skin lesion classification.

In International Conference on Medical Image

Computing and Computer-Assisted Intervention (pp.

336-346). Cham: Springer International Publishing.

Chowdhury, A., Kassem, H., Padoy, N., Umeton, R., &

Karargyris, A. 2021. A review of medical federated

learning: Applications in oncology and cancer research.

In International MICCAI Brainlesion Workshop (pp. 3-

24). Cham: Springer International Publishing.

Deng, X., Oda, S., Kawano, Y. 2023. Graphene-based

midinfrared photodetector with bull’s eye plasmonic

antenna. Optical Engineering, 62(9), 097102-097102.

Duarte, A. F., Sousa-Pinto, B., Azevedo, L. F., Barros, A.

M., Puig, S., Malvehy, J., ... & Correia, O. 2021.

Clinical ABCDE rule for early melanoma detection.

European Journal of Dermatology, 31(6), 771-778.

Hameed, S. S., Hassan, W. H., Latiff, L. A., & Ghabban, F.

2021. A systematic review of security and privacy

issues in the internet of medical things; the role of

machine learning approaches. PeerJ Computer Science,

7, e414.

Hashmani, Manzoor Ahmed, et al. 2021. An adaptive

federated machine learning-based intelligent system for

skin disease detection: A step toward an intelligent

dermoscopy device. Applied Sciences, 11(5), 2145.

Hekler, A., Maron, R. C., Haggenmüller, S., Schmitt, M.,

Wies, C., Utikal, J. S., ... & Brinker, T. J. 2024. Using

multiple real-world dermoscopic photographs of one

lesion improves melanoma classification via deep

learning. Journal of the American Academy of

Dermatology.

Hosseini, S. M., Sikaroudi, M., Babaie, M., & Tizhoosh, H.

R. 2023. Proportionally fair hospital collaborations in

federated learning of histopathology images. IEEE

transactions on medical imaging.

Liakos, K. G., Busato, P., Moshou, D., Pearson, S., &

Bochtis, D. 2018. Machine learning in agriculture: A

review. Sensors, 18(8), 2674.

Liu, Y., Liu, L., Yang, L., Hao, L. and Bao, Y. 2021.

Measuring distance using ultra-wideband radio

technology enhanced by extreme gradient boosting

decision tree (XGBoost). Automation in Construction,

126, 103678.

Liu, Y. and Bao, Y. 2023. Intelligent monitoring of

spatially-distributed cracks using distributed fiber optic

sensors assisted by deep learning. Measurement, 220,

113418.

Qiu, Y., Wang, J., Jin, Z., Chen, H., Zhang, M., & Guo, L.

2022. Pose-guided matching based on deep learning for

assessing quality of action on rehabilitation training.

Biomedical Signal Processing and Control, 72, 103323.

Qiu, Y., Yang, Y., Lin, Z., Chen, P., Luo, Y., & Huang, W.

2020. Improved denoising autoencoder for maritime

image denoising and semantic segmentation of USV.

China Communications, 17(3), 46-57.

Riaz, S., Naeem, A., Malik, H., Naqvi, R. A., & Loh, W. K.

2023. Federated and Transfer Learning Methods for the

Classification of Melanoma and Nonmelanoma Skin

Cancers: A Prospective Study. Sensors, 23(20), 8457.

Roschewitz, David, et al. 2021. Ifedavg: Interpretable data-

interoperability for federated learning. arXiv preprint

arXiv:2107.06580.

Shi, Yujun, et al. 2023. Understanding and mitigating

dimensional collapse in federated learning. IEEE

Transactions on Pattern Analysis and Machine

Intelligence.

Sugaya, T., Deng, X. 2019. Resonant frequency tuning of

terahertz plasmonic structures based on solid

immersion method. 2019 44th International Conference

on Infrared, Millimeter, and Terahertz Waves, 1-2.

Tammina, S. 2019. Transfer learning using vgg-16 with

deep convolutional neural network for classifying

images. International Journal of Scientific and Research

Publications (IJSRP), 9(10), 143-150.

Wicaksana, J., Yan, Z., Yang, X., Liu, Y., Fan, L., & Cheng,

K. T. 2022. Customized federated learning for multi-

source decentralized medical image classification.

IEEE Journal of Biomedical and Health Informatics,

26(11), 5596-5607.

Xu, G., Wu, Y., Hu, J., & Shi, Y. 2022. Achieving fairness

in dermatological disease diagnosis through automatic

weight adjusting federated learning and personalization.

arXiv preprint arXiv:2208.11187.

Yaqoob, M. M., Alsulami, M., Khan, M. A., Alsadie, D.,

Saudagar, A. K. J., AlKhathami, M., & Khattak, U. F.

2023. Symmetry in privacy-based healthcare: a review

of skin cancer detection and classification using

federated learning. Symmetry, 15(7), 1369.

Yu, Z., Dong, Y., Cheng, J., Sun, M., & Su, F. 2022.

Research on face recognition classification based on

improved GoogleNet. Security and Communication

Networks, 2022, 1-6.

Yuan, Z. W., & Zhang, J. 2016. Feature extraction and

image retrieval based on AlexNet. In Eighth

International Conference on Digital Image Processing

(ICDIP 2016) (Vol. 10033, pp. 65-69). SPIE.

Zhao, F., Yu, F., Trull, T., & Shang, Y. 2023. A new

method using LLMs for keypoints generation in

qualitative data analysis. In 2023 IEEE Conference on

Artificial Intelligence (CAI) (pp. 333-334). IEEE.

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

678