US-20260123902-A1 - MEDICAL IMAGE PROCESSING APPARATUS, MEDICAL IMAGE LEARNING METHOD, AND MEDICAL IMAGE PROCESSING METHOD

US20260123902A1US 20260123902 A1US20260123902 A1US 20260123902A1US-20260123902-A1

Abstract

A method for training a medical image performed by a medical image processing apparatus for processing a medical image for a body is provided. The method for training a medical image includes preparing a first input data set including a training chest X-ray image, and a bone enhancement image or a bone extraction image acquired from the training chest X-ray image; preparing a label data for the first input data set including osteoporosis information or bone mineral density information corresponding to the training chest X-ray image; and training an artificial neural network model using the first input data set and the label data.

Inventors

Namkug Kim
Miso JANG
Min Gyu Kim

Assignees

PROMEDIUS INC.

Dates

Publication Date: 20260507
Application Date: 20260102
Priority Date: 20200909

Claims (10)

1 . A method for training an artificial neural network model by using a medical image, to be performed by a medical image processing apparatus, the method comprising: configuring, as a first input data set, a training chest X-ray image; configuring, as first label data for the first input data set, osteoporosis information or bone mineral density information corresponding to the training chest X-ray image; configuring, as a second input training data set, a fragment image including only some of a plurality of bones included in the training chest X-ray image; configuring, as second label data, the osteoporosis information or the bone mineral density information corresponding to the second input training data set; and training an artificial neural network model using the first input data set, the first label data, the second input training data set and the second label data, wherein the first input data set contains more information than the second input training data set, and wherein the fragment image is an image including two or less of a clavicle, cervical vertebrae, thoracic vertebrae, and scapula for the training chest X-ray image.
2 . The method of claim 1 , wherein the configuring the first input data set includes configuring a soft tissue image acquired from the training chest X-ray image, and the first input data set further includes the soft tissue image.
3 . The method of claim 1 , wherein the configuring the first input data set includes configuring a bone enhancement image or a bone extraction image acquired from the training chest X-ray image, and the first input data set further includes the bone enhancement image or the bone extraction image.
4 . The method of claim 1 , comprising: pre-training the artificial neural network model using the second input training data set, and then post-training the artificial neural network model using the first input data set.
5 . A method for processing a medical image to be performed by a medical image processing apparatus, the method comprising: inputting a chest X-ray image of a diagnosis target to an artificial neural network model, wherein the artificial neural network model is trained by: configuring, as a first input data set, a training chest X-ray image; configuring, as first label data for the first input data set, osteoporosis information or bone mineral density information corresponding to the training chest X-ray image; configuring, as a second input training data set, a fragment image including only some of a plurality of bones included in the training chest X-ray image; configuring, as second label data, the osteoporosis information or the bone mineral density information corresponding to the second input training data set; training an artificial neural network model using the first input data set, the first label data, the second input training data set and the second label data; and outputting osteoporosis information or bone mineral density information acquired from the chest X-ray image of the diagnosis target using the artificial neural network model, wherein the first input data set contains more information than the second input training data set, and wherein the fragment image is an image including two or less of a clavicle, cervical vertebrae, thoracic vertebrae, and scapula for the training chest X-ray image.
6 . The method of claim 5 , wherein the outputting the osteoporosis information or bone mineral density information includes outputting a class activation map (CAM) in which the chest X-ray images of the diagnosis target are classified into a normal group and a risk group for osteoporosis on the basis of the osteoporosis information or the bone mineral density information.
7 . The method of claim 5 , wherein the configuring the first input data set includes configuring a soft tissue image acquired from the training chest X-ray image, and the first input data set further includes the soft tissue image.
8 . The method of claim 5 , wherein the configuring the first input data set includes configuring a bone enhancement image or a bone extraction image acquired from the training chest X-ray image, and the first input data set further includes the bone enhancement image or the bone extraction image.
9 . The method of claim 5 , wherein the training the artificial neural network model includes pre-training the artificial neural network model using the second input training data set, and then post-training the artificial neural network model using the first input data set.
10 . A non-transitory computer-readable storage medium storing a computer program thereon, the medium comprising instructions for controlling a processor to perform a method for processing a medical image, the method comprising: inputting a chest X-ray image of a diagnosis target to an artificial neural network model, wherein the artificial neural network model is trained by: configuring, as a first input data set, a training chest X-ray image; configuring, as a first label data for the first input data set, osteoporosis information or bone mineral density information corresponding to the training chest X-ray image; configuring, as a second input training data set, a fragment image including only some of a plurality of bones included in the training chest X-ray image; configuring, as second label data, the osteoporosis information or the bone mineral density information corresponding to the second input training data set; and training an artificial neural network model using the first input data set, the first label data, the second input training data set and the second label data; and outputting osteoporosis information or bone mineral density information acquired from the chest X-ray image of the diagnosis target by the artificial neural network model, wherein the first input data set contains more information than the second input training data set, and wherein the fragment image is an image including two or less of a clavicle, cervical vertebrae, thoracic vertebrae, and scapula for the training

Description

CROSS-REFERENCE TO RELATED APPLICATIONS The present application is a continuation of U.S. application Ser. No. 18/264,976, filed on Aug. 10, 2023, which is the U.S. National Stage of International Patent Application No. PCT/KR2021/012271, filed on Sep. 9, 2021, which is based upon and claims the benefit of priority to Korean Patent Application No. 10-2020-0115689, filed Sep. 9, 2020, the entire contents of which are incorporated herein for all purposes by this reference. TECHNICAL FIELDS The present invention relates to a method and apparatus for learning medical images of a body, and to a method and apparatus for processing the medical images of the body. BACKGROUND ART A medical imaging apparatus is equipment for acquiring an image of an internal structure of a body of a diagnosis target. The medical imaging apparatus is a non-invasive inspection apparatus, which photographs and processes structural details, internal tissues, and fluid flows in the body and shows these to a user. The user such as a doctor may diagnose a health condition and disease of a patient using a medical image output from the medical imaging apparatus. Examples of the medical imaging apparatus include an X-ray imaging apparatus that irradiates a target with X-rays and detects X-rays passing through the target to capture an image, a magnetic resonance imaging (MRI) apparatus for providing a magnetic resonance image, a computed tomography (CT) apparatus, and an ultrasound diagnostic apparatus, and among these, an X-ray imaging apparatus is most widely used. Meanwhile, according to “Choi, H. J., et al., Burden of osteoporosis in adults in Korea: a national health insurance database study. Journal of bone and mineral metabolism, 2012. 30 (1): p. 54-58.” and “Lee, Y.-K., B.-H. Yoon, and K.-H. Koo, Epidemiology of osteoporosis and osteoporotic fracture in South Korea. Endocrinology and Metabolism, 2013. 28 (2): p. 90-93.”, Korea entered an aged society in which a proportion of people aged 65 and older is 14% in 2018, and is expected to enter a super-aged society in which the proportion of the people aged 65 and older is over 25% in 2025. The aged society and/or the super-aged society causes various aging problems; one of the problems is that the number of osteoporosis patients increases, and medical costs and economic costs for a fracture due to the osteoporosis and the osteoporosis rapidly increase. The osteoporosis refers to a condition in which bone strength is lowered and a fracture is highly likely to occur, is a bone disease that progresses throughout the entire body, and is a skeletal disease in which the risk of the fracture increases due to a loss of bone strength. According to “Hong, S. and K. Han, The incidence of hip fracture and mortality rate after hip fracture in Korea: A nationwide population-based cohort study. Osteoporosis and Sarcopenia, 2019.” and “Jung, H.-S., et al., Incidence of Osteoporotic Refractures Following Proximal Humerus Fractures in Adults Aged 50 Years and Older in Korea. Journal of Bone Metabolism, 2019. 26 (2): p. 105-111.”, osteoporosis has no symptoms, but when a fracture occurs, a secondary fracture is highly likely to occur and the occurrence of complications increases, and thus, it is important to prevent a bone mineral density from being reduced before the fracture occurs, and to perform a screening test for reduction in the bone mineral density in order to prevent the bone mineral density from being reduced. Bone strength is determined by bone quality and bone mineral density, which are determined by bone turnover rate, structure, fine damage, and mineralization, and since 80% of the bone strength depends on the bone mineral density, bone mineral density measurement is a useful method for osteoporosis diagnosis. According to “LewieckiEM, LaneNE. Common mistakes in the clinical use of bone mineral density testing. Nat Clin Pract Rheumatol.2008; 4:667-674”, “Jae Gyoon Kim and Young Wan Moon. 2011. Diagnosis of Osteoporosis. Hip & Pelvis (formerly Journal of the Korean Hip Society), 23 (2): 108-115”, and “Ho-Sung, K., Tae-Hyung, K., & Sang-Hyun, K. (2018). Management Methods of Bone Mineral Density Examination Using Dual Energy X-ray Absorptiometry. Journal of Radiological Science and Technology, 41 (4), 351-360”, the International Society for Clinical Densitometry (ISCD), an international academic organization, recognizes dual energy X-ray absorptiometry (DXA) as the most suitable bone mineral density measurement method. With the dual energy X-ray absorptiometry, bone mineral densities of a lumbar spine and a femur, which are central bones, are measured and diagnosis into any one of osteoporosis, osteopenia, and normality is performed on the basis of a smallest value thereof. According to a study using the National Health and Nutrition Examination Survey published in “Lee, K.-S., et al., New reference data on bone mineral density and the prevalence of osteoporosis in Korean adults aged 50 years or