1: \begin{abstract}

2: Federated learning and its application to medical image segmentation have recently become a popular research topic. This training paradigm suffers from statistical heterogeneity between participating institutions' local datasets, incurring convergence slowdown as well as potential accuracy loss compared to classical training. To mitigate this effect, federated personalization emerged as the federated optimization of one model per institution. We propose a novel personalization algorithm tailored to the feature shift induced by the usage of different scanners and acquisition parameters by different institutions. 

3: %Moreover, the proposed method is the first to account for both inter and intra-institution feature shift (multiple scanners used in a single institution). 

4: This method is the first to account for both inter and intra-institution feature shift (multiple scanners used in a single institution).

5: %Our method is based on the computation of radiomic features on each volume followed by a clustering in the feature space server-side. 

6: It is based on the computation, within each centre, of a series of radiomic features capturing the global texture of each 3D image volume, followed by a clustering analysis pooling all feature vectors transferred from the local institutions to the central server. 

7: %We propose to finetune a global model obtained through classical federated learning on each computed clustered decentralized datasets. 

8: Each computed clustered decentralized dataset (potentially including data from different institutions) then serves to finetune a global model obtained through classical federated learning.

9: We validate our approach on the Federated Brain Tumor Segmentation 2022 Challenge dataset (FeTS2022). Our code is available at (\url{https://github.com/MatthisManthe/radiomics_CFFL}).

10: 

11: \end{abstract}

12: