US-12622636-B2 - Body fluid volume estimation device, body fluid volume estimation method, and non-transitory computer-readable medium

US12622636B2US 12622636 B2US12622636 B2US 12622636B2US-12622636-B2

Abstract

A body fluid volume estimation device includes a pre-training unit, a transfer learning unit, and an estimation unit. The pre-training unit performs pre-training by using, as supervised information, information indicating body fluid volumes of the multiple patients when face images of multiple patients are captured. The transfer learning unit further performs transfer learning on multiple face images of one specific patient after the pre-training, and constructs a trained model. The estimation unit estimates, by inputting a face image of the one specific patient to the trained model, a body fluid volume at a point in time at which the face image of the one specific patient is captured. By estimating a body fluid volume from a face image by machine learning, the body fluid volume can be used for assistance such as decision making of a user.

Inventors

Yusuke AKAMATSU
Yoshifumi Onishi
Hideo Tsurushima

Assignees

NEC CORPORATION

Dates

Publication Date: 20260512
Application Date: 20230630
Priority Date: 20220707

Claims (11)

1 . A body fluid volume estimation device comprising: at least one memory configured to store instructions; and at least one processor configured to execute the instructions to: perform pre-training on face images of multiple patients by using, as supervised information, information indicating a body fluid volume of each of multiple patients when the face images of multiple patients are captured; perform transfer learning on multiple face images of one specific patient after the pre-training, and construct a trained model; and estimate, by inputting a face image of the one specific patient to the trained model, a body fluid volume of the one specific patient at a point in time at which the face image of the one specific patient is captured, wherein the at least one processor is further configured to execute the instructions to perform pre-training by weight-aware supervised momentum contrast (WeightSupMoCo).
2 . The body fluid volume estimation device according to claim 1 , wherein the information indicating the body fluid volume of each multiple patients when the face images of the multiple patients are captured includes information indicating presence or absence of a swelling in the face image of each of the multiple patients, and information indicating weight of each of the multiple patients, and wherein the at least one processor is further configured to execute the instructions to estimate presence or absence of a swelling and predicts weight of the one specific patient at a point in time at which the face image of the one specific patient is captured by inputting the face image of the one specific patient to the trained model.
3 . The body fluid volume estimation device according to claim 2 , wherein the at least one processor is further configured to execute the instructions to: detect, based on an prediction result of the presence or absence of the swelling of the one specific patient, whether the body fluid volume of the one specific patient is changed from a preset standard body fluid volume of the one specific patient; and acquire, from an estimation result of the weight, a difference in the body fluid volume of the one specific patient from the standard body fluid volume.
4 . The body fluid volume estimation device according to claim 2 , wherein the information indicating the presence or absence of the swelling in the face image of each of the multiple patients is label information representing the presence or absence of the swelling, and wherein the at least one processor is further configured to execute the instructions to perform pre-training in such a way that feature values of face images having the same label information representing the presence or absence of the swelling are brought closer as the information indicating the weight is more similar.
5 . The body fluid volume estimation device according to claim 2 , wherein the multiple patients and the one specific patient are a patient who receives dialysis, and wherein weight of each of the multiple patients and the one specific patient after dialysis associated with a case without a swelling has a value acquired by subtracting a body fluid volume removed by dialysis from weight of each of the multiple patients and the one specific patient before dialysis associated with a case with a swelling.
6 . The body fluid volume estimation device according to claim 2 , wherein the multiple patients and the one specific patient are a patient who receives dialysis, and wherein weight of each of the multiple patients and the one specific patient before dialysis associated with a case with a swelling has a value acquired by adding a body fluid volume removed by dialysis to weight of each of the multiple patients and the one specific patient after dialysis associated with a case without a swelling.
7 . The body fluid volume estimation device according to claim 2 , wherein the multiple patients and the one specific patient are a patient who receives dialysis, and wherein weight of each of the multiple patients and the one specific patient before dialysis associated with a case with a swelling has a value acquired by adding a body fluid volume removed by dialysis to preset standard weight of each of the multiple patients and the one specific patient.
8 . The body fluid volume estimation device according to claim 1 , wherein the at least one memory is further configured to store the face images of the multiple patients to be used for pre-training, the information indicating the body fluid volume of the multiple patients when the face images of the multiple patients are captured, and the multiple face images of the one specific patient to be used for the transfer learning, wherein the at least one processor is further configured to execute the instructions to: read, from the at least one memory, the face images of the multiple patients and information indicating the body fluid volume when the face images of the multiple patients are captured, and performs pre-training; read, from the at least one memory, the multiple face images of the one specific patient; and perform transfer learning.
9 . The body fluid volume estimation device according to claim 1 , further comprising an image capturing device, wherein the face image of the one specific patient is captured by the image capturing device.
10 . A body fluid volume estimation method performed by at least one processor of a body fluid volume estimation device including the at least one processor and at least one memory configured to store instructions, the processor executing the instructions to: perform pre-training on face images of multiple patients by using, as supervised information, information indicating a body fluid volume of each of the multiple patients when the face images of the multiple patients are captured; perform transfer learning on multiple face images of one specific patient after the pretraining, and constructing a trained model; and estimate, by inputting a face image of the one specific patient to the trained model, a body fluid volume of the one specific patient at a point in time at which the face image of the one specific patient is captured, wherein the at least one processor is further configured to execute the instructions to perform pre-training by weight-aware supervised momentum contrast (WeightSupMoCo).
11 . A non-transitory computer-readable medium storing a program causing at least one processor of a body fluid volume estimation device including the at least one processor and at least one memory configured to store instructions to execute the instructions to: perform pre-training on face images of multiple patients by using, as supervised information, information indicating a body fluid volume of each of the multiple patients when the face images of the multiple patients are captured; perform transfer learning on multiple face images of one specific patient after the pretraining, and constructing a trained model; and estimate by inputting a face image of the one specific patient to the trained model, a body fluid volume of the one specific patient at a point in time at which the face image of the one specific patient is captured, wherein the at least one processor is further configured to execute the instructions to perform pre-training by weight-aware supervised momentum contrast (WeightSupMoCo).

Description

INCORPORATION BY REFERENCE This application is based upon and claims the benefit of priority from Japanese patent application No. 2022-109664, filed on Jul. 7, 2022, the disclosure of which is incorporated herein in its entirety by reference. TECHNICAL FIELD The present disclosure relates to a body fluid volume estimation device, a body fluid volume estimation method, and a non-transitory computer-readable medium. BACKGROUND ART In recent years, development of a technique for determining a health state of a person and presence or absence of a disease from an appearance of the person, for example, an image of a face and the like has been progressing (Published Japanese Translation of PCT International Publication for Patent Application, No. 2022-512044, Japanese Unexamined Patent Application Publication No. 2020-199072, and Japanese Unexamined Patent Application Publication No. 2005-65812). In general, it is known that a change occurs in a form of a face and a lower limb according to a health state of a person. An increase in capacity of a body is detected as a swelling, and a decrease in capacity is detected as a decrease in firmness of skin. A swelling mainly refers to a state where excessive water is accumulated in a gap of tissue and a body fluid volume increases, and occurs by various causes such as a central disease, a respiratory/circulatory disease, a renal disease, an orthopedic disease, a metabolic disease, and a malignant tumor. Further, a decrease in firmness of skin occurs by a decrease in water in a body, and occurs in various states such as dehydration and heatstroke. For example, a swelling occurs when a waste matter and water in a body cannot be removed due to a decrease in renal function, and thus excessive water in a body is removed by dialysis treatment in a present condition. Therefore, it is important for a dialysis patient to maintain water in a body (i.e., a body fluid volume) within a desirable range, and thus an intake of water and salt needs to be limited in daily life. As a change in weight, factors such as a change in body fluid volume, fat mass, and muscle mass are conceivable. Since it is conceivable that dialysis is performed for approximately four hours, and a change in fat mass and muscle mass does not occur during the dialysis, a change in weight by the dialysis conceivably reflects a change in body fluid volume. Further, for elderly patients having a chronic heart failure, it is also inconceivable that muscle mass is increased by exercise and an increase in amount of food leads to an increase in fat, and thus a change in weight may conceivably represent a change in body fluid volume. Furthermore, also in dehydration, fat mass and muscle mass do not change in a short period of time, and thus a change in weight may conceivably represent a change in body fluid volume. In order to recognize a state of a patient having a disease accompanied by a swelling, a degree of the swelling of the patient, i.e., a body fluid volume is required to be measured. As a general swelling estimation technique, for example, a technique of measuring a degree of a swelling by a pitting test by a medical staff member has been proposed (J. Chen et. al, “Camera-Based Peripheral Edema Measurement Using Machine Learning,” in Proc. IEEE Int. Conf. Healthcare Informatics (ICHI), 2018, pp. 115-122.) In this technique, an image is captured during a pitting test of a lower limb, and a degree of a peripheral swelling of a lower limb or the like is estimated from the image by support vector machine (SVM) or a convolutional neural network (CNN). Further, a technique (A. G. Smith et. al, “Objective determination of peripheral edema in heart failure patients using short-wave infrared molecular chemical imaging,” Journal of Biomedical Optics, vol. 26, No. 10, pp. 105002, 2021) of measuring a degree of a swelling, based on an image in which a peripheral portion such as hands and feet of a patient is captured, by using a short-wave infrared (SWIR) camera has been proposed. In this technique, by using a property in which an absorption coefficient of water, collagen, and fat increases in a specific spectrum region of the SWIR camera in presence of a swelling, a swelling level can be estimated from a spectrum component. SUMMARY As described above, in order to manage a degree of a swelling of a patient, i.e., a body fluid volume on a daily basis, it is desirable that the patient himself/herself measures a body fluid volume, and manages an intake of water according to a measurement result. However, the technique for measuring a body fluid volume described above needs to be performed by a professional medical staff member, and there is also a restriction that specific equipment such as a special camera is needed. Meanwhile, in order for a patient to autonomously limit an intake of water and salt in daily life, a technique for a patient to be able to measure a body fluid volume on a daily basis is required to be established