METHOD AND APPARATUS FOR TRAINING A COMPUTER-IMPLEMENTED MACHINE-LEARNING MODEL

    公开(公告)号:EP4372617A1

    公开(公告)日:2024-05-22

    申请号:EP23169568.5

    申请日:2023-04-24

    摘要: A method for training a computer-implemented machine-learning model f for controlling a system by control parameter y, wherein said model has been trained to predict a target variable y = f(x) based on an input feature value X, where X is an input value feature value obtained by one or more sensors and used to generate by said model said target variable y as control parameter to thereby control said system,
    wherein said model f has been trained using a first set of training data (yk, xk) corresponding to k =1... N time points to find the trained machine learning model f*
    wherein a backward model g* has been trained based on at least a part of said first set of training data (yk, xk) to predict an input feature value xk+1 based on its preceding input feature value xk and the corresponding target variable yk, said method comprising:
    using said backward model g* to generate second set of training data, which comprises the future input feature values xk+1 predicted by the backward machine learning model g* and the corresponding ground truth variable yk+1,
    training said model f, which has been trained using said first set of training data, using said replay buffer as said second set of training data.

    AI-ASSISTED GENERATION OF ANNOTATED MEDICAL IMAGES

    公开(公告)号:EP4239589A1

    公开(公告)日:2023-09-06

    申请号:EP22159493.0

    申请日:2022-03-01

    摘要: A method for AI-assisted generation of annotated medical images (WSI) is described. In the method, a set (S) of pre-annotated medical images (WSI) is received. Then the received set (S) is processed by automatically performing the steps of training an AI-based uncertainty model (BSM) using the received set (S) as training data. Then, a step of processing medical images (WSI) of the received set (S) by determining classified segments (SG) and/or uncertainty regions (UR) in the medical images (WSI) using the trained AI-based uncertainty model (BSM) follows. Further, a step of selecting at least a part of the processed medical images (WSI) including classified segments and/or uncertainty regions (UR) based on the processing result and presenting the selected part of processed medical images (WSI) to a human expert (HE) is performed. Furthermore, a modified received set (S') including additional annotations (AA) created by the human expert (HE) is received. The mentioned steps can be repeated using the modified set (S') of medical images (WSI) as training data, until a predetermined quality criteria for the annotated medical images (WSI) is achieved. Further, an annotation assistance device (10) is described. Furthermore, a computer-implemented method for providing a segmented medical image is described.

    MASS PRODUCTION OF HUMAN PLURIPOTENT STEM CELL DERIVED CARDIAC STROMAL CELL

    公开(公告)号:EP3945133A8

    公开(公告)日:2023-04-12

    申请号:EP20188364.2

    申请日:2020-07-29

    IPC分类号: C12N5/077 C12N5/071

    摘要: The application describes a method for producing a population of cardiac stromal cells from pluripotent stem cells. Specifically, the method relates to (i) inducing epithelial-mesenchymal transition of pluripotent stem cell derived epicardial cells and (ii) amplifying the number of cardiac stromal cells in serum-free conditions. These cardiac stromal cells can be mass produced according to the described method and said cells maintain the expression of CD90, CD73 and CD44 in at least 80% of the cardiac stromal cells. Furthermore, the application relates to a population of cardiac stromal cells, which are pluripotent stem cells derived and wherein at least 80% of the cardiac stromal cells express CD90, CD73 and CD44. Said cardiac stromal form the basis for several in vitro and in vivo applications such as the production of engineered organ tissue and the support of, for example, heart repair. Also, a serum-free culture medium for the amplification of cardiac stromal cells is provided herein.

    Pouch-like structure with paracrine activity and method for its preparation
    7.
    发明公开
    Pouch-like structure with paracrine activity and method for its preparation 审中-公开
    Beutelartige Struktur mitParakrinaktivitätund Verfahren zu deren Herstellung

    公开(公告)号:EP2842581A1

    公开(公告)日:2015-03-04

    申请号:EP13182437.7

    申请日:2013-08-30

    摘要: The present invention relates to a pouch-like structure useful for mechanically preventing distension and/or resisting dilation of the heart and for supporting the hearts function by controllable and paracrine support of a failing heart in a mammal. The pouch-like structure is composed at least partly of engineered tissue comprising genetically engineered cells, like, genetically engineered cells other than cardiac myocytes whereby said genetically engineered cells, like genetically engineered cells other than cardiac myocytes, contain a gene encoding a paracrine factor said gene encoding the paracrine factor being under control of an inducible promoter system or an heterologous promoter system. Further, the present invention relates to a method for the preparation and individual of the pouch-like structure for therapeutic, disease modelling, and drug development applications. In addition, the present invention relates to cells other than cardiac myocytes for use in the preparation of the pouch-like structure as described herein.

    摘要翻译: 本发明涉及一种用于机械地防止心脏膨胀和/或抵抗心脏扩张并且用于通过哺乳动物中失败的心脏的可控和旁分泌支持来支持心脏功能的袋状结构。 袋状结构至少部分地由包含基因工程细胞的工程化组织组成,如心肌细胞以外的遗传工程细胞,其中所述基因工程细胞,如心肌细胞以外的基因工程细胞,含有编码旁分泌因子的基因, 编码旁分泌因子的基因在诱导型启动子系统或异源启动子系统的控制下。 此外,本发明涉及用于治疗,疾病建模和药物开发应用的袋状结构的制备和个体的方法。 此外,本发明涉及用于制备如本文所述的袋状结构的心肌细胞以外的细胞。