Customized Federated Learning for Multi-Source Decentralized Medical Image Classification

Abstract

The performance of deep networks for medical image analysis is often constrained by limited medical data, which is privacy-sensitive. Federated learning (FL) alleviates the constraint by allowing different institutions to collaboratively train a federated model without sharing data. However, the federated model is often suboptimal with respect to the characteristics of each client's local data. Instead of training a single global model, we propose Customized FL (CusFL), for which each client iteratively trains a client-specific/private model based on a federated global model aggregated from all private models trained in the immediate previous iteration. Two overarching strategies employed by CusFL lead to its superior performance: 1) the federated model is mainly for feature alignment and thus only consists of feature extraction layers; 2) the federated feature extractor is used to guide the training of each private model. In that way, CusFL allows each client to selectively learn useful knowledge from the federated model to improve its personalized model. We evaluated CusFL on multi-source medical image datasets for the identification of clinically significant prostate cancer and the classification of skin lesions.

Improving Pancreatic Cyst Management: Artificial Intelligence-Powered Prediction of Advanced Neoplasms through Endoscopic Ultrasound-Guided Confocal Endomicroscopy

“…ML models are then trained using these centralized data. In contrast, FL aims to train AI models while keeping the data decentralized at multiple data collection sites [70,71]. In the mainstream FL framework, a centralized server coordinates the model training process.…”

“…The server aggregates these client model parameters into a single ML model. This approach enables collaborative model development and adaptation among multiple hospitals without the need to share any patient data [71][72][73][74][75][76][77]. FL has demonstrated reliable results when applied to clinical information, PET images, and multimodal data.…”

Enhanced Deep Learning Approach for Accurate Eczema and Psoriasis Skin Detection

“…They introduced a new dataset called “DermaNet”, which consists of 23 diseases; however, they obtained a very low accuracy of 67% for the diagnosis of the skin diseases from this dataset. Other lightweight deep learning methods for melanoma detection are proposed [ 27 , 28 , 29 , 30 , 31 , 32 ], such as the method presented by Biasi et al [ 27 ]. The authors put forward a proposition for the creation and execution of a hybrid architecture that draws upon Cloud, Fog, and Edge Computing.…”

Medical Imaging Applications of Federated Learning

“…Wicaksana et al proposed Customized FL (CusFL) and demonstrated its ability to detect prostate cancer and identify skin cancer [89]. This group's approach differed from traditional FL by iteratively training a client-specific model based on the global model instead of training a single one, thereby avoiding catastrophic forgetting.…”

“…Additionally, Sarma et al conducted a case study demonstrating their ability to utilize FL to train a deep learning model across three academic institutions while preserving patients' privacy [92]. Furthermore, Wicaksana et al, as mentioned in the skin section, proposed a new way to train the FL model and demonstrated the abilities of the said model on prostate cancer data [89].…”