公开(公告)号：US20220189009A1

公开(公告)日：2022-06-16

申请号：US17123967

申请日：2020-12-16

Applicant: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

Inventor： Ting-Yuan WANG ,  Ming-Shan DENG ,  Ya-Wen LEE ,  Jung-Tzu LIU

IPC: G06T7/00 ,  G06K9/62 ,  G06T3/40 ,  G06T7/73 ,  G06K9/20 ,  G06T7/60 ,  A61B5/00 ,  A61B3/12 ,  G06N20/00 ,  G16H30/40

Abstract: A medical image analysis method includes: reading an original medical image; performing image classification and object detection on the original medical image to generate a first classification result and a plurality of object detection results by a plurality of complementary artificial intelligence (AI) models; performing object feature integration and transformation on a first detection result and a second detection result among the object detection results to generate a transformation result by a features integration and transformation module; and performing machine learning on the first classification result and the transformation result to generate an image interpretation result by a machine learning module and display the image interpretation result.

公开(公告)号：US20210201086A1

公开(公告)日：2021-07-01

申请号：US16904161

申请日：2020-06-17

Applicant: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

Inventor： Po-Yi WU ,  Ming-Shan DENG

IPC: G06K9/62 ,  G06K9/32

Abstract: A training system and method of object detection model is disclosed. The training system includes an object detection model and a loss calculation module. The object detection model is configured to generate an output image according to an input image. The loss calculation module, coupled to the object detection model, is configured to calculate a total classification loss value according to the output image and a solution image, calculate a loss value according to the total classification loss value, and transmit the loss value to the object detection model. The total classification loss value is calculated according to a number of classification losses corresponding to a number of object types. Each classification loss corresponding to each object type is calculated according to a first parameter, a second parameter and a third parameter.