System und Verfahren zur Automatisierten Domänenanpassung in der Medizinischen Bildgebung

Details

Veröffentlichungs-Nr.

EP4804051

Anmeldetag

3. März 2025

Veröffentlichung

9. September 2026

Rechtsraum

EP

Abstract

The invention relates to a system (200) and a method (300) for automated Domain Adaptation, hereafter "DA", determination for a Deep Learning, hereafter "DL", model trained on a training dataset of a source domain, hereafter "D<S>", and that has to be applied on a target dataset of a target domain, hereafter "D<T>", wherein the DL model is characterized by an encoder-decoder architecture, wherein the "trained" DL model refers hereafter to the DL model trained on D<S>, wherein D<S> and D<T> differ in that they present a different distribution of data, the method (300) comprising: - determining (302) if labels are available for D<T>, i.e. if one or several labels are associated to one or several data of D<T>, wherein each label associated to a data is configured for providing a context to that data for enabling the DL model to learn from it, enabling thus to train the DL model on the labeled data; - if labels are available for D<T>, then automatically launching(303) an evaluation of N supervised domain adaptation, hereafter "SDA", methods, with N ≥ 2, wherein said evaluation of the N SDA methods comprises an iterative testing process configured for testing the N SDA methods on a first test dataset of D<T>, and automatically selecting the best-performing SDA method among said N SDA methods for the DA of the trained DL model with respect to D<T>, wherein the first test dataset comprises at least part of the labeled data of D<T>; - otherwise, if labels are not available for D<T>, then automatically selecting (304) an unsupervised domain adaptation, hereafter "UDA" for handling said target dataset.

Anmelder

Firmen

Centre Hospitalier Universitaire Vaudois
Siemens Healthineers AG

Land

🇨🇭 Schweiz