Title:
LEARNING DEVICE, METHOD, AND PROGRAM, AND MEDICAL IMAGE PROCESSING DEVICE
Document Type and Number:
WIPO Patent Application WO/2022/138277
Kind Code:
A1
Abstract:
Provided are a learning device, a method, and a program, and a medical image processing device, all of which make it possible to create a learning model having a high reliability of disease detection and reduce annotation cost for preparing a dataset for learning. A first processor of this learning device reads first medical images (a1) having disease labels (x) from a dataset (B) stored in a memory (14) and inputs the first medical images to a first learning model (16). The first medical images (a1) are normalized on the basis of a lung region inferred by the first learning model (16). By using the normalized first medical images (A1) and the disease labels (x), an untrained second learning model (18) for detecting a disease is trained. When this second learning model (18) is to be trained, uncertainty values (C1) of the first medical images (a1) are calculated on the basis of uncertainties (c1) inferred simultaneously by the first learning model (16), and a first medical image (a1) having a high uncertainty value (C1) is excluded from learning data.
Inventors:
HIASA YUTA (JP)
Application Number:
PCT/JP2021/045790
Publication Date:
June 30, 2022
Filing Date:
December 13, 2021
Export Citation:
Assignee:
FUJIFILM CORP (JP)
International Classes:
G06T7/00; A61B6/00; G06N20/00
Foreign References:
| JP2020086519A | 2020-06-04 | |||
| JP2020503603A | 2020-01-30 | |||
| JP2019149093A | 2019-09-05 | |||
| US20190122073A1 | 2019-04-25 | |||
| JP2020091862A | 2020-06-11 |
Attorney, Agent or Firm:
MATSUURA, Kenzo (JP)
Download PDF: