Citation

Mahmud, Mamun and Rafi, Najmus Sakib and Uddain, Rayhan and Goh, Kah Ong Michael and Fahad, Nafiz and Rabbi, Riadul Islam and Tusher, Ekramul Haque and Tarin, Fatema Mostafa and Connie, Tee (2026) Privacy-Preserving Federated Deep Learning for Multi-Class Liver Fibrosis Staging from Ultrasound Images. In: 2026 9th International Conference on Artificial Intelligence and Big Data (ICAIBD), 29-31 May 2026, Chengdu, China.

Text 3.pdf - Published Version Restricted to Repository staff only Download (598kB)

Abstract

A privacy-preserving federated deep learning framework for multiclass liver-fibrosis staging from ultrasound images is presented in this paper. 20% of the samples in the dataset (Kaggle) were kept as an independent test set, and the remaining 80% were divided equally between two simulated clients in order to simulate a two-site deployment. Federated Averaging (FedAvg) was used to aggregate identical EfficientNet-B7 backbones that had been trained locally. The held-out test set (n=1,264) was used to assess the optimal global checkpoint chosen throughout the federated rounds. With class-wise F1-scores ranging from 0.97 to 1.00, the federated EfficientNet-B7 model achieved 98.81% accuracy with macro-averaged precision =98.44%, recall =98.32%,andF1-score =98.38%. These findings suggest that federated training with privacy preservation can get close to centralized performance while preserving data locality. Future research will confirm the methodology on diverse datasets and investigate improved privacy, personalization, and communication efficiency strategies.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Federated Learning, Privacy-Preserving Ma chine Learning, EfficientNet-B7, Liver Fibrosis, Ultrasound Imaging, Multi-class Classification, Federated Averaging (Fe dAvg), Medical Image Analysis
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	02 Oct 2026 02:16
Last Modified:	02 Oct 2026 02:16
URII:	http://shdl.mmu.edu.my/id/eprint/16824

Downloads

Edit (login required)