TWI779963B - Nutritional status assessment method and nutritional status assessment system - Google Patents

Abstract

A nutritional state assessment method, comprising: training and generating a muscle segmentation model and a fat segmentation model based on multiple abdominal medical images for training, multiple muscle area labeling data and multiple fat area labeling data; Area, fat area area for multiple training and prognostic nutritional index category for multiple training, training generates an index classification model; for an abdominal medical image to be analyzed, use the muscle segmentation model and the fat segmentation model to segment a muscle area image and a fat region image; calculating the area of the muscle region image and the fat region image; and for a muscle region area and a fat region area, using the index classification model, to generate an indicator indicative of one of a plurality of prognostic nutritional index categories index classification results.

Description

Nutritional status assessment method and nutritional status assessment system

The present invention relates to an assessment method using artificial intelligence technology, in particular to a nutritional status assessment method. The present invention also relates to a nutritional status assessment system.

Malnutrition is quite a fatal problem for cancer patients. Malnutrition impairs organ function, leading to an increased risk of infection and decreased survival. At present, the most commonly used assessment of the overall nutritional status of patients is through blood drawing, and one of the main analysis items is the nutritional index (prognostic nutrition index, PNI).

Blood drawing to test a patient's nutritional status is an invasive test. When intensive monitoring of nutritional status is required, patients will not be willing to test due to the uncomfortable feeling of the blood drawing process. In addition, the above-mentioned blood drawing values are often affected by the severity of the patient's co-morbidity and non-tumor-related inflammation at the time, which makes the values fluctuate greatly, which is not conducive to accurately assessing the patient's health status.

Another conventional method for evaluating a patient's nutritional status is based on the physician's judgment based on the patient's abdominal medical images. However, this method relies heavily on professional physicians, but due to the limited manpower of physicians, it cannot be popularized and used clinically, and human judgment still has subjective elements and may make mistakes.

How to develop a new nutritional status assessment method that can improve the above-mentioned shortcomings of the prior art is a subject to be further explored in the present invention.

Therefore, the object of the present invention is to provide a method for evaluating nutritional status.

Another object of the present invention is to provide a nutritional status assessment system.

Therefore, the nutritional status evaluation method of the present invention is implemented by a nutritional status evaluation system. The method includes: training the An original muscle segmentation model, which generates a muscle segmentation model, wherein each muscle area label data indicates the area showing muscles in the corresponding abdominal medical image for training; according to the abdominal medical image for training and multiple records corresponding to The fat area labeling data of the abdominal medical image for training is used to train an original fat segmentation model to generate a fat segmentation model, wherein each fat area labeling data indicates the area showing fat in the corresponding abdominal medical image for training; According to multiple training muscle area areas, multiple training fat area areas corresponding to the training muscle area areas, and multiple training training area areas corresponding to the training muscle area areas and the training fat area areas training an original index classification model with prognostic nutritional index categories, generating an index classification model, wherein each trained prognostic nutritional index category indicates one of a plurality of prognostic nutritional index categories; for an abdominal medical image to be analyzed, using the A muscle segmentation model, segmenting a muscle region image from the abdominal medical image to be analyzed; for the abdominal medical image to be analyzed, using the fat segmentation model to segment a fat region image from the abdominal medical image to be analyzed; calculating the calculating the area of the muscle region image to generate a muscle region area; calculating the area of the fat region image to generate a fat region area; and using the index classification model for the muscle region area and the fat region area to generate an index classification result, The index classification result indicates one of the prognostic nutritional index categories.

In some implementation aspects, the nutritional status assessment method further includes: according to multiple training images of the third lumbar abdominal abdomen, multiple training non-third lumbar abdominal medical images, and multiple records corresponding to the training An original image classification model is trained by using the third lumbar abdominal medical images and the image category labeling data of the non-third lumbar abdominal medical images for training, and an image classification model is generated, wherein, corresponding to the training third lumbar abdominal The image category labeling data of the medical images indicate a third lumbar spine image category, and the image category labeling data corresponding to the training non-third lumbar spine abdominal medical images indicate a non-third lumbar spine image category; for a patient's A plurality of abdominal medical images, using the image classification model to generate a plurality of image classification results respectively corresponding to the abdominal medical images, each image classification result indicating the third lumbar image category or the non-third lumbar image category; and according to As for the image classification results, one of the abdominal medical images is selected as the abdominal medical image to be analyzed, wherein the image classification result corresponding to the abdominal medical image to be analyzed indicates the third lumbar image category.

In some implementations, the training abdominal medical images, the training third lumbar abdominal medical images and the abdominal medical images are images generated by computerized tomography.

In some implementation aspects, when generating the index classification model, it is also based on multiple training heights respectively corresponding to the muscle area for training, the fat area for training, and the prognostic nutritional index categories for training , multiple training ages and multiple training genders, and train the original index classification model. When generating the index classification result, the index classification model is also used for a height, an age and a gender.

The nutritional status evaluation system of the present invention includes a storage unit and a processing unit. The storage unit stores multiple abdominal medical images for training, multiple pieces of muscle area labeling data respectively corresponding to the training abdominal medical images, multiple pieces of fat area labeling data respectively corresponding to the training abdominal medical images, multiple The area of muscle area for training, the area of fat area for training respectively corresponding to the area of muscle area for training, and the prognosis for training respectively corresponding to the area of muscle area for training and the area of fat area for training Nutrition index category, wherein, each muscle area labeling data indicates the area showing muscle in the corresponding abdominal medical image for training, and each fat area labeling data indicates the corresponding fat area showing in the abdominal medical image for training, Each training prognostic nutritional index category indicates one of a plurality of prognostic nutritional index categories. The processing unit is electrically connected to the storage unit.

The processing unit trains an original muscle segmentation model to generate a muscle segmentation model according to the abdominal medical images for training and the muscle region labeling data.

The processing unit trains an original fat segmentation model to generate a fat segmentation model according to the abdominal medical images for training and the fat region labeling data.

The processing unit trains an original index classification model according to the muscle area areas for training, the fat area areas for training and the prognostic nutritional index categories for training to generate an index classification model.

For an abdominal medical image to be analyzed, the processing unit uses the muscle segmentation model to segment a muscle area image from the abdominal medical image to be analyzed.

For the abdominal medical image to be analyzed, the processing unit uses the fat segmentation model to segment a fat region image from the abdominal medical image to be analyzed.

The processing unit calculates the area of the muscle region image to generate a muscle region area.

The processing unit calculates the area of the fat region image to generate a fat region area.

For the muscle region area and the fat region area, the processing unit uses the index classification model to generate an index classification result indicating one of the prognostic nutritional index categories.

The efficacy of the present invention lies in: the muscle segmentation model is generated by training according to the training abdominal medical images and the muscle region labeling data, and is generated by training according to the training abdominal medical images and the fat region labeling data The fat segmentation model generates the index classification model by training the muscle area for training, the fat area for training, and the prognostic nutritional index category for training, and by analyzing the abdominal medical image Using the muscle segmentation model and the fat segmentation model to separately segment the muscle region image and the fat region image from the abdominal medical image to be analyzed, and by using the index classification model for the muscle region area and the fat region area The classification results of this index can avoid the uncomfortable feeling caused by the previous method of blood drawing to test the nutritional status of patients, and the situation that the value of the blood test value is affected by the inflammation situation. The efficiency and accuracy of the evaluation are greatly improved, and the nutritional status evaluation can be carried out for patients who have obtained abdominal medical images without additional examination.

Before the present invention is described in detail, it should be noted that in the following description, similar elements are denoted by the same numerals.

Referring to FIG. 1 , an embodiment of the nutritional status assessment method of the present invention is implemented by a nutritional status assessment system 100 , and the nutritional status assessment system 100 includes a storage unit 1 and a processing unit 2 .

The storage unit 1 stores a plurality of third lumbar abdominal medical images for training, a plurality of non-third lumbar abdominal medical images for training, and multiple records corresponding to the training third lumbar abdominal medical images and the training non-third Image category labeling data of the third lumbar abdominal medical images, multiple training abdominal medical images, multiple pieces of muscle region labeling data corresponding to the training abdominal medical images, multiple pieces of training abdominal medical images respectively corresponding Fat area labeling data, multiple training muscle area areas, multiple training fat area areas corresponding to the training muscle areas, and multiple training muscle area areas corresponding to the training muscle areas Prognostic nutritional index (PNI) categories for training on fat area area.

The image category labeling data corresponding to the third lumbar abdominal medical images for training indicates a third lumbar image category, and the image category labeling data corresponding to the training non-third lumbar abdominal medical images indicate a non- Third lumbar image category.

Each muscle area labeling data indicates a muscle area in the corresponding abdominal medical image for training, and each fat area labeling data indicates a fat area in the corresponding abdominal medical image for training. Each training prognostic nutritional index category indicates one of a plurality of prognostic nutritional index categories.

The image category labeling data, the muscle region labeling data and the fat region labeling data are, for example, produced by professional physicians. The prognostic nutritional index categories for training are, for example, generated by classifying the prognostic nutritional indices obtained from blood tests.

The storage unit 1 can be implemented using a non-volatile storage medium such as a hard disk, flash memory, and the like.

The processing unit 2 is electrically connected to the storage unit 1 . The processing unit 2 may include (but not limited to) a single-core processor, a multi-core processor, a dual-core mobile phone processor, a microprocessor, a microcontroller, a digital signal processor (DSP), a field programmable At least one of a Programmable Gate Array (FPGA), an Application Specific Integrated Circuit (ASIC), and a Radio Frequency Integrated Circuit (RFIC).

Referring to FIG. 1 and FIG. 2 (including FIG. 2A and FIG. 2B ), the steps of the nutritional status assessment method of the present invention will be described below. First, as shown in step S01, the processing unit 2 trains an original image classification model according to the training medical images of the third lumbar abdomen, the training non-third lumbar abdominal medical images, and the image category labeling data , generating an image classification model. In this embodiment, the original image classification model uses ResNet101 as the training framework.

Next, as shown in step S02 , the processing unit 2 trains an original muscle segmentation model according to the abdominal medical images for training and the muscle region labeling data to generate a muscle segmentation model. In this embodiment, the original muscle segmentation model uses Unet++ as the training framework.

Next, as shown in step S03 , the processing unit 2 trains an original fat segmentation model according to the training abdominal medical images and the fat region labeling data to generate a fat segmentation model. In this embodiment, the original fat segmentation model uses Unet++ as the training framework.

Next, as shown in step S04, the processing unit 2 trains an original index classification model according to the training muscle area, the training fat area and the training prognostic nutritional index category to generate an index classification model . In this embodiment, the original exponential classification model uses a 5-layer fully connected network (Fully connected network, FCN) as a training framework.

In this embodiment, when generating the index classification model, the processing unit 2 is also based on a plurality of records respectively corresponding to the muscle area for training, the fat area for training, and the prognostic nutritional index categories for training. The height for training, the age for multiple training and the gender for multiple training are used to train the original index classification model.

The execution order of steps S01 to S04 can be changed arbitrarily, and is not limited to the order of this embodiment.

Next, as shown in step S05, for a plurality of abdominal medical images of a patient, the processing unit 2 uses the image classification model to generate a plurality of image classification results respectively corresponding to the abdominal medical images, and each image classification result indicates The third lumbar image category or the non-third lumbar image category.

Next, as shown in step S06, the processing unit 2 selects one of the abdominal medical images as an abdominal medical image to be analyzed according to the image classification results, wherein the abdominal medical image to be analyzed corresponds to the The image classification result indicates the third lumbar image category.

Next, as shown in step S07 , for the abdominal medical image to be analyzed, the processing unit 2 uses the muscle segmentation model to segment a muscle region image from the abdominal medical image to be analyzed.

Next, as shown in step S08 , for the abdominal medical image to be analyzed, the processing unit 2 uses the fat segmentation model to segment a fat area image from the abdominal medical image to be analyzed.

The order of steps S07 and S08 can be reversed, and is not limited to the order of this embodiment.

Next, as shown in step S09 , the processing unit 2 calculates the area of the muscle region image to generate a muscle region area. Step S09 may be executed after step S07 and before step S11, and the sequence of this embodiment is not limited.

Next, as shown in step S10 , the processing unit 2 calculates the area of the fat region image to generate a fat region area. Step S10 may be executed after step S08 and before step S11, and the order of this embodiment is not limited.

Next, as shown in step S11, for the muscle region area and the fat region area, the processing unit 2 uses the index classification model to generate an index classification result indicating one of the prognostic nutritional index categories. In this embodiment, the number of the prognostic nutritional index categories is two, respectively representing the category of PNI<50 and the category of PNI>=50, but not limited thereto. In addition, in this embodiment, when generating the index classification result, the processing unit 2 also uses the index classification model for a height, an age and a gender.

It is added that the training abdominal medical images, the training third lumbar abdominal medical images and the abdominal medical images are images generated by computerized tomography (CT), but not limited thereto. The training abdominal medical images, the training third lumbar abdominal medical images and the abdominal medical images may also be images produced by magnetic resonance imaging (MRI), for example.

It is worth mentioning that the present invention can effectively improve the accuracy of the index classification results by selecting the abdominal medical images that include the third lumbar vertebrae. However, the abdominal medical images selected by the present invention are not limited to the third lumbar vertebrae. The image model is not limited to binary classification. In other implementations, it is also possible to select the second and fourth lumbar abdominal medical images. The multi-class classification models were selected separately.

In summary, the nutritional status assessment method of the present invention generates the muscle segmentation model by training according to the abdominal medical images for training and the muscle region labeling data, and generates the muscle segmentation model based on the abdominal medical images for training and the fat The fat segmentation model is generated by region labeling data training, and the index classification model is generated by training according to the muscle region area for training, the fat region area for training and the prognostic nutritional index category for training, and by for the The abdominal medical image to be analyzed uses the muscle segmentation model and the fat segmentation model to separately segment the muscle region image and the fat region image from the abdominal medical image to be analyzed, and by calculating the muscle region and the fat region Using the index classification model to generate the index classification results can avoid the uncomfortable feeling caused by the previous method of blood drawing to test the nutritional status of patients, and the situation that the blood test value is affected by inflammation. Interpreting the shortcomings of images can greatly improve the efficiency and accuracy of evaluation, and for patients who have obtained abdominal medical images, nutritional status evaluation can be performed without additional examination, so the purpose of the present invention can indeed be achieved.

But the above-mentioned ones are only embodiments of the present invention, and should not limit the scope of the present invention. All simple equivalent changes and modifications made according to the patent scope of the present invention and the content of the patent specification are still within the scope of the present invention. Within the scope covered by the patent of the present invention.

100: Nutritional Status Assessment System 1: storage unit 2: Processing unit S01~S12: Steps

Other features and effects of the present invention will be clearly presented in the implementation manner with reference to the drawings, wherein: Fig. 1 is a schematic diagram of a hardware connection relationship of an embodiment of the nutritional status assessment method of the present invention; and FIG. 2 (including FIG. 2A and FIG. 2B ) is a flowchart of the embodiment.

S01~S12: Steps

Claims (8)

A nutritional status assessment method implemented by a nutritional status assessment system, the method comprising: According to a plurality of abdominal medical images for training and a plurality of muscle region labeling data respectively corresponding to the abdominal medical images for training, an original muscle segmentation model is trained to generate a muscle segmentation model, wherein each muscle region labeling data indicates The region showing muscles in the corresponding abdominal medical image for training; According to the abdominal medical images for training and a plurality of fat region labeling data respectively corresponding to the abdominal medical images for training, an original fat segmentation model is trained to generate a fat segmentation model, wherein each fat region labeling data indicates Areas showing fat in the corresponding abdominal medical images for training; According to multiple training muscle area areas, multiple training fat area areas corresponding to the training muscle area areas, and multiple training training area areas corresponding to the training muscle area areas and the training fat area areas training a raw index classification model with prognostic nutritional index categories, generating an index classification model, wherein each trained prognostic nutritional index category indicates one of a plurality of prognostic nutritional index categories; For an abdominal medical image to be analyzed, use the muscle segmentation model to segment a muscle region image from the abdominal medical image to be analyzed; For the abdominal medical image to be analyzed, using the fat segmentation model to segment a fat region image from the abdominal medical image to be analyzed; calculating the area of the muscle region image to generate a muscle region area; calculating the area of the fat region image to generate a fat region area; and Using the index classification model for the muscle region area and the fat region area, an index classification result is generated, the index classification result indicating one of the prognostic nutritional index categories. The nutritional status assessment method as described in claim 1, further comprising: According to multiple medical images of the third lumbar abdomen for training, multiple medical images of the non-third lumbar abdomen for training, and multiple records corresponding to the medical images of the third lumbar abdomen for training and the non-third lumbar abdomen for training Image category labeling data of medical images, training an original image classification model to generate an image classification model, wherein the image category labeling data corresponding to the training third lumbar spine abdomen medical images indicate a third lumbar spine image category, The image category annotation data corresponding to the training non-third lumbar abdominal medical images indicates a non-third lumbar image category; For multiple abdominal medical images of a patient, using the image classification model to generate a plurality of image classification results respectively corresponding to the abdominal medical images, each image classification result indicates the third lumbar image category or the non-third lumbar image category; and According to the image classification results, one of the abdominal medical images is selected as the abdominal medical image to be analyzed, wherein the image classification result corresponding to the abdominal medical image to be analyzed indicates the third lumbar image category. The nutritional status assessment method as described in Claim 2, wherein the training abdominal medical images, the training third lumbar abdominal medical images, and the abdominal medical images are images generated by computerized tomography. The nutritional status assessment method as described in claim 2, wherein, when generating the index classification model, it is also based on the area corresponding to the muscle area for training, the area of fat area for training, and the prognostic nutritional index for training The multiple training heights, multiple training ages and multiple training genders of the category are used to train the original index classification model; When generating the index classification result, the index classification model is also used for a height, an age and a gender. A nutritional status assessment system comprising: A storage unit that stores multiple abdominal medical images for training, multiple pieces of muscle area labeling data respectively corresponding to the training abdominal medical images, multiple pieces of fat area labeling data respectively corresponding to the training abdominal medical images, Multiple training muscle area areas, multiple training fat area areas corresponding to the training muscle area areas, and multiple training training fat area areas corresponding to the training muscle area areas and the training fat area areas The category of prognostic nutritional index, wherein each muscle area labeling data indicates the area showing muscle in the corresponding abdominal medical image for training, and each fat area labeling data indicates the area showing fat in the corresponding training abdominal medical image , each prognostic nutritional index class for training indicates one of a plurality of prognostic nutritional index classes; and a processing unit electrically connected to the storage unit; The processing unit trains an original muscle segmentation model to generate a muscle segmentation model according to the abdominal medical images used for training and the muscle region labeling data; The processing unit trains an original fat segmentation model to generate a fat segmentation model according to the abdominal medical images used for training and the fat region labeling data; The processing unit trains an original index classification model to generate an index classification model according to the training muscle region area, the training fat region area and the training prognostic nutritional index category; For an abdominal medical image to be analyzed, the processing unit uses the muscle segmentation model to segment a muscle region image from the abdominal medical image to be analyzed; For the abdominal medical image to be analyzed, the processing unit uses the fat segmentation model to segment a fat region image from the abdominal medical image to be analyzed; The processing unit calculates the area of the muscle region image to generate a muscle region area; The processing unit calculates the area of the fat region image to generate a fat region area; For the muscle region area and the fat region area, the processing unit uses the index classification model to generate an index classification result indicating one of the prognostic nutritional index categories. The nutritional status assessment system according to claim 5, wherein the storage unit also stores a plurality of third lumbar abdominal medical images for training, a plurality of non-third lumbar abdominal medical images for training, and multiple records corresponding to the The image category labeling data of the third lumbar abdominal medical image for training and the image category labeling data of the non-third lumbar abdominal medical image for training, wherein the image category labeling data corresponding to the third lumbar abdominal medical image for training indicates one The third lumbar image category, corresponding to the image category annotation data of the non-third lumbar abdominal medical images for training indicates a non-third lumbar image category; The processing unit trains an original image classification model to generate an image classification model according to the third lumbar abdominal medical images for training, the non-third lumbar abdominal medical images for training, and the image category labeling data; For a plurality of abdominal medical images of a patient, the processing unit uses the image classification model to generate a plurality of image classification results respectively corresponding to the abdominal medical images, each image classification result indicates the third lumbar image category or the non-third lumbar image category Three lumbar image categories; The processing unit selects one of the abdominal medical images as the abdominal medical image to be analyzed according to the image classification results, wherein the image classification result corresponding to the abdominal medical image to be analyzed indicates the third lumbar image kind. The nutritional status evaluation system as described in claim 6, wherein the training abdominal medical images, the training third lumbar abdominal medical images, and the abdominal medical images are images generated by computerized tomography. The nutritional status assessment system as described in Claim 6, wherein, when generating the index classification model, it is also based on the areas corresponding to the muscle area for training, the area of fat area for training, and the prognostic nutritional index for training The multiple training heights, multiple training ages and multiple training genders of the category are used to train the original index classification model; When generating the index classification result, the index classification model is also used for a height, an age and a gender.

Images