WO2020224282A1 - 一种婴幼儿粪便采样图像分类处理系统及方法 - Google Patents

一种婴幼儿粪便采样图像分类处理系统及方法 Download PDF

Info

Publication number
WO2020224282A1
WO2020224282A1 PCT/CN2019/130562 CN2019130562W WO2020224282A1 WO 2020224282 A1 WO2020224282 A1 WO 2020224282A1 CN 2019130562 W CN2019130562 W CN 2019130562W WO 2020224282 A1 WO2020224282 A1 WO 2020224282A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
classification
infant
images
subsystem
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/130562
Other languages
English (en)
French (fr)
Inventor
秦文健
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Publication of WO2020224282A1 publication Critical patent/WO2020224282A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Definitions

  • the invention relates to a computer system that can realize data statistics and processing in the process of infant disease screening, in particular to an infant stool sampling image classification processing system and method.
  • biliary atresia is the most serious digestive tract disease in infants and young children.
  • jaundice is the first symptom. When the jaundice appears, the stool also becomes light yellow. As the jaundice worsens, it gradually becomes white or even clay-like gray. Therefore, the color of infant stool images has become an important basis for judging the disease.
  • Existing screening methods include: screening using a stool color chart and screening according to the saturation (S) value in the HSV model of the stool image.
  • biliary atresia is the most serious digestive tract disease in infants and young children, accounting for half of obstructive jaundice in infants and young children.
  • the incidence rate is 1:8000-1:14000 surviving infants and young children, but there are big differences in region and ethnicity. Most cases have been reported in Asia. The incidence of Eastern peoples is 4-5 times higher, and the ratio of men to women is 1:2.
  • Congenital biliary atresia is a kind of obstruction of the bile ducts inside and outside the liver, which can lead to cholestatic liver cirrhosis. Congenital biliary atresia and eventually liver failure.
  • the existing early screening techniques include stool colorimetric cards and screening based on the saturation (S) value in the HSV model of stool image color.
  • S saturation
  • the deficiencies of using the stool color chart are: First, there are errors in the conclusions drawn by the parents after comparing the color of the color chart. According to reports in the literature, the parents of children who compare the color of the color chart think that normal stool is actually Abnormal biliary obstruction in stool, this error rate is as high as 30%.
  • Second, medical staff at all levels are not very aware of and vigilant about biliary atresia. At present, there are a large number of pediatric patients in mainland China, and doctors are scarce. It is impossible to use limited manpower and material resources in screening for biliary atresia and other diseases. The results.
  • the technical problem to be solved by the present invention is that in view of the shortcomings of the prior art, it provides a method for extracting relevant feature information from infant feces images, and using deep learning methods to classify the images, thereby improving the accuracy of judgment and recognition.
  • a sampling image classification processing system and method that help reduce screening errors.
  • the present invention adopts the following technical solutions.
  • An infant feces sampling image classification processing system which includes: an image acquisition subsystem for recording infant feces images; an image classification subsystem for classification training on infant feces images entered by the image acquisition subsystem, Get the classification training result.
  • the images recorded by the image acquisition subsystem include infant feces images taken by the user through the image recording device and child feces images collected from the hospital.
  • the image classification subsystem includes: a target area acquisition unit for acquiring a target area in an infant feces image; an image cropping unit for cropping an image of a preset size in the target area; feature extraction The unit is used to perform feature extraction on the cropped image; the classification training unit is used to perform classification training on infant feces images using a preset classification model according to the extracted image features, and obtain classification training results.
  • the infant feces image input by the image acquisition subsystem is an image with a pixel size of 256*256
  • the image cropping unit is used to crop an image with a pixel size of 224*224 in the target area.
  • An image classification and processing method for infant feces sampling is implemented based on a system.
  • the system includes an image acquisition subsystem and an image classification subsystem.
  • the method includes: step S1, using the image acquisition subsystem to record infants Infant feces images; step S2, using the image classification subsystem to perform classification training on the infant feces images entered by the image acquisition subsystem to obtain classification training results.
  • the image classification subsystem includes a target area acquisition unit, an image cropping unit, a feature extraction unit, and a classification training unit.
  • the execution process of step S2 includes: step S20, using the target area acquisition unit to acquire infants The target area in the stool image; step S21, using the image cropping unit to crop an image of a preset size in the target area; step S22, using the feature extraction unit to perform feature extraction on the cropped image; step S23
  • the classification training unit uses a preset classification model to perform classification training on the infant feces images to obtain classification training results.
  • the infant feces image entered by the image acquisition subsystem is an image with a pixel size of 256*256
  • the image cropping unit is configured to be in the target area Cut out an image with a pixel size of 224*224.
  • the classification model includes the VGGNet16 model, the ResNet50 model, and the Inception V3 model in GoogleNet, and in step S23, the VGGNet16 model, the ResNet50 model and the Inception V3 model in GoogleNet are used for classification training.
  • the classification model in step S23 is a convolutional neural network.
  • the image features extracted in the step S22 include texture features and color features of the image.
  • the infant feces sampling image classification and processing system disclosed in the present invention the infant feces images are first obtained, relevant color, texture and other characteristic information are extracted from the early infant feces images, and then the images are classified and trained based on the deep learning method, and then The classification training result is obtained.
  • the present invention greatly improves the accuracy of image classification and recognition, greatly improves the efficiency of disease screening, and provides a better aid for the screening of early infant disease. effect.
  • Figure 1 is a block diagram of the sampled image classification processing system of the present invention
  • Fig. 2 is a flowchart of a method for classifying and processing sampled images according to the present invention.
  • Fig. 1 discloses an image classification and processing system for infant feces sampling, please refer to Fig. 1, which includes:
  • Image acquisition subsystem 1 used to record images of infant feces
  • the image classification subsystem 2 is used to perform classification training on the infant feces images entered by the image acquisition subsystem 1 to obtain classification training results.
  • the infant feces image is first obtained, the relevant color, texture and other characteristic information are extracted from the early infant feces image, and then the image is classified and trained based on the deep learning method to obtain the classification training result.
  • the present invention greatly improves the accuracy of image classification and recognition, greatly improves the efficiency of disease screening, and plays a better auxiliary role for the screening of early infant disease.
  • the images recorded by the image acquisition subsystem 1 include infant feces images taken by the user through an image recording device and child feces images collected from the hospital.
  • the image classification subsystem 2 includes:
  • the target area acquiring unit 20 is configured to acquire the target area in the infant feces image
  • the image cropping unit 21 is configured to crop an image of a preset size in the target area
  • the feature extraction unit 22 is configured to perform feature extraction on the cropped image
  • the classification training unit 23 is configured to perform classification training on the infant feces images by using a preset classification model according to the extracted image features to obtain classification training results.
  • the infant feces image entered by the image acquisition subsystem 1 is an image with a pixel size of 256*256, and the image cropping unit 21 is used to crop an image with a pixel size of 224*224 in the target area .
  • the present invention also discloses a method for classification and processing of infant feces sampling images. As shown in FIG. 1 and FIG. 2, the method is implemented based on a system that includes image acquisition Subsystem 1 and Image Classification Subsystem 2, the method includes:
  • Step S1 using the image acquisition subsystem 1 to record infant feces images
  • Step S2 using the image classification subsystem 2 to perform classification training on the infant feces images entered by the image acquisition subsystem 1 to obtain classification training results.
  • the image classification subsystem 2 includes a target area acquisition unit 20, an image cropping unit 21, a feature extraction unit 22, and a classification training unit 23.
  • the execution process of step S2 includes:
  • Step S20 using the target area acquiring unit 20 to acquire a target area in an infant feces image
  • Step S21 using the image cropping unit 21 to crop an image of a preset size in the target area;
  • Step S22 using the feature extraction unit 22 to perform feature extraction on the cropped image
  • step S23 the classification training unit 23 uses a preset classification model to perform classification training on the infant feces images according to the extracted image features to obtain classification training results.
  • the infant feces image entered by the image acquisition subsystem 1 is an image with a pixel size of 256*256
  • the image cropping unit 21 is used to display the image on the target
  • An image with a pixel size of 224*224 is cropped from the area.
  • the principle of data enhancement is to adopt some techniques on the basis of the original data. Means to generate new data. Due to the characteristics of fecal image data, it is the first choice to obtain the region of interest in the data before data enhancement.
  • the method of random cropping is used. The result of cropping is to randomly crop the fecal image with the input size of 256*256 into an image with the size of 224*224. Distribute a large number of infant feces data in the virtual cloud to achieve efficient and safe storage of image data and high-speed dynamic read and write. Then, feature extraction is performed on the obtained stool image data.
  • the classification model includes the VGGNet16 model, the ResNet50 model, and the Inception V3 model in GoogleNet.
  • the VGGNet16 model, the ResNet50 model and the Inception V3 model in GoogleNet are used for classification training. Furthermore, during the training process, the parameters of the model are fine-tuned to obtain a higher classification accuracy. After the three classification models are fine-tuned, the accuracy of the training process can reach more than 95%. In the test results, its accuracy rate can also reach more than 90%.
  • the classification model in step S23 is a convolutional neural network. That is, the convolutional neural network is used to train and test the data, and then obtain the classification result.
  • two classification methods of machine learning such as support vector machine and K nearest neighbor algorithm, can also be used to classify fecal image data.
  • the image features extracted in the step S22 include texture features and color features of the image.
  • the system and method for classification and processing of infant feces sampling images disclosed in the present invention can be applied to a variety of disease screening activities based on analysis and processing of infant fecal image data.
  • the following takes infant early biliary atresia disease screening as an example.
  • the technical scheme of the present invention is illustrated by examples:
  • the processing process of the infant feces sampling image classification and processing system in this embodiment is as follows: first use the image acquisition subsystem 1 to record infant feces images, extract relevant color, texture and other characteristic information from the early infant feces images, and then use The image classification subsystem 2 performs classification training on the infant feces images entered by the image acquisition subsystem 1, and then obtains the classification training results.
  • the above processing process effectively improves the accuracy of image classification and recognition, and can improve the screening of early biliary atresia.
  • the inspection efficiency plays a good auxiliary role in the screening of early infant biliary atresia diseases.
  • the images recorded by the image acquisition subsystem 1 include infant feces images taken by a user through an image recording device and fecal images of children with biliary atresia collected from the hospital.
  • the experimental data collection is obtained by the doctor or the patient's family members taking pictures of the infant feces images, and then the doctors calibrating the feces images (normal, abnormal).
  • two doctors jointly calibrate the data. When the two doctors have the same calibration results on the same stool image, they can distinguish which type the image belongs to. When the calibration results of two doctors on the same stool image are inconsistent, we ignore this image and exclude it from the data set.
  • the specific process includes: establishing a large sample database of fecal pictures of normal children and children with biliary atresia.
  • Different mobile phone models and different lighting conditions have an important impact on image analysis, and the increasing number of image data, data storage and calculation are also very large challenge.
  • image enhancement algorithms research on cloud computing-based medical imaging information storage and analysis platform, a large number of infant fecal data and diagnostic information are distributed in the virtual cloud, to achieve efficient and safe storage of image data and high-speed dynamic read and write; research is based on cloud platform Parallel computing of data to realize the parallelization of feature learning algorithms and learning classification algorithms, develop a complete set of mobile application client terminals, realize information sharing, and patient’s family members upload the data to the “cloud” storage to obtain the pre-judgment result. The patient can also pass The cloud platform provides expert consultation services.
  • intelligent recognition technology that automatically recognizes abnormal fecal pictures is adopted, and the computer automatically quantifies diagnostic indicators of infant feces images to assist doctors in completing the diagnosis of diseases more accurately and safely, and research and use deep learning features
  • the learning method extracts image features, and studies the use of weak supervision to build a classifier to classify images.
  • Evaluate the value of network intelligent recognition technology in the early diagnosis of biliary atresia collect large samples of clinical cases in the early stage, and develop a mobile application client. In the actual application process, it can be combined with multiple medical centers for treating biliary atresia and multiple maternal and child health care.
  • the center participates in the project to collect data samples, which can continuously and dynamically obtain the development of lesions, and use the client to follow-up and follow up the cases, and finally evaluate and verify the reliability and accuracy of the intelligent identification algorithm.
  • the system and method for classifying and processing infant feces sampling images disclosed in the present invention use deep learning methods to screen for diseases such as early biliary atresia, can fully extract the features of fecal images, and avoid the use of fecal colorimetric cards. Human error because of the impact.
  • the present invention can integrate the deep learning algorithm into the mobile phone APP software, and automatically recognize normal or abnormal fecal images by taking pictures of the mobile phone, which greatly improves the screening efficiency of diseases such as early biliary atresia. More accurate and safe assistant doctors complete the diagnosis of diseases.
  • the present invention uses a deep learning method for screening, first extracts the features of the stool image and then performs training tests, which greatly improves the accuracy of classification and recognition.
  • infant stool sampling image classification and processing system and method proposed in the present invention are actually computer systems and computer methods used to realize data statistics and processing in the infant disease screening process, and are not related to a specific disease. Treatment and screening methods.
  • the present invention does not limit specific application scenarios, that is, whether the present invention is used as an auxiliary computer solution for screening infants and young children, it should belong to the protection scope of the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Quality & Reliability (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Image Analysis (AREA)
  • Investigating Or Analysing Biological Materials (AREA)

Abstract

一种婴幼儿粪便采样图像分类处理系统，其包括有：图像采集子系统（1），用于录入婴幼儿粪便图像；图像分类子系统（2），用于对图像采集子系统录入的婴幼儿粪便图像进行分类训练，得出分类训练结果。上述采样图像分类处理系统中，首先获取婴幼儿粪便图像，从早期婴幼儿粪便图像中提取相关颜色、纹理等特征信息，再基于深度学习的方式对图像进行分类训练，进而得到分类训练结果，大大提高了图像分类识别的准确率和疾病筛查的效率，为早期婴幼儿疾病的筛查工作起到了较好的辅助作用。

Description

一种婴幼儿粪便采样图像分类处理系统及方法 技术领域
本发明涉及婴幼儿疾病筛查过程中可实现数据统计和处理的计算机系统，尤其涉及一种婴幼儿粪便采样图像分类处理系统及方法。
背景技术
现有技术中，在一些婴幼儿疾病筛查过程中，通常会涉及一些数据统计和处理的方法，以早期婴幼儿胆道闭锁疾病筛查为例：胆道闭锁是最严重的婴幼儿消化道疾病，临床上以黄疸为首发症状，在黄疸出现的同时，粪便亦变成淡黄色，随黄疸加重逐渐趋于偏白甚至陶土样灰白色，因此婴幼儿粪便图像的颜色成为了判别该疾病的重要依据。现有的筛查方法有：使用粪便比色卡进行筛查和根据粪便图像的HSV模型中的饱和度(S)的值来进行筛查。
目前，胆道闭锁是最严重的婴幼儿消化道疾病，占婴幼儿梗阻性黄疸疾病的一半。其发病率为1∶8000-1∶14000个存活出生婴幼儿，但地区和种族有较大差异。以亚洲报道的病例为多，东方民族的发病率高4-5倍，男女之比为1∶2。先天性胆道闭锁是一种肝内外胆管出现阻塞，并可导致淤胆性肝硬化先天性胆道闭锁而最终发生肝功能衰竭，是小儿外科领域中最重要的消化外科疾病之一，也是小儿肝移植中最常见的适应证。胆道闭锁的典型病例婴幼儿为足月产，大多数并无异常，粪便色泽正常，黄疸一般在生后2-3周逐渐显露，有些病例的黄疸出现于生后最初几天当时误诊为生理性黄疸。粪便变成棕黄、淡黄米色，以后成为无胆汁的陶土样灰白色。
现有的早期筛查技术包括粪便比色卡和基于粪便图像颜色HSV模型中饱和度(S)的值进行筛查。其中使用粪便比色卡的不足之处在于：一是家长对照比色卡颜色后得出的结论存在误差，据文献报道，患儿家长通过对照比色卡颜色后认为正常的粪便，其实仍是不正常的胆道梗阻粪便，这种失误率高达30％。二是各级医务人员对胆道闭锁的认识和警惕性不高，目前国内大陆地区儿科患儿人数众多，医生稀缺，无法用有限的人力和物力在胆道闭锁等疾病的筛查工作中取得较好的成绩。
由此可见，现有图像数据统计和处理方法中，使用粪便图像的HSV模型中饱和度(S)值来进行筛查存在着不足之处：一是单以HSV模型中的饱和度(S)的值来进行筛查会出现一定的误差，二是识别判断的准确率不高。
发明内容
本发明要解决的技术问题在于，针对现有技术的不足，提供一种从婴幼儿粪便图像中提取相关特征信息，利用深度学习的方法对图像进行分类训练，进而提高判断识别的准确率、有助于减少筛查误差的采样图像分类处理系统及方法。
为解决上述技术问题，本发明采用如下技术方案。
一种婴幼儿粪便采样图像分类处理系统，其包括有：图像采集子系统，用于录入婴幼儿粪便图像；图像分类子系统，用于对图像采集子系统录入的婴幼儿粪便图像进行分类训练，得出分类训练结果。
优选地，所述图像采集子系统录入的图像包括用户通过图像摄录装置拍摄的婴幼儿粪便图像以及从医院收集的患儿粪便图像。
优选地，所述图像分类子系统包括：目标区域获取单元，用于获取婴幼儿粪便图像中的目标区域；图像裁剪单元，用于在所述目标区域中裁剪出预设尺寸的图像；特征提取单元，用于对裁剪得到的图像进行特征提取；分类训练单元，用于根据所提取的图像特征，利用预设的分类模型对婴幼儿粪便图像进行分类训练，得出分类训练结果。
优选地，所述图像采集子系统录入的婴幼儿粪便图像是像素大小为256*256的图像，所述图像裁剪单元用于在所述目标区域中裁剪出像素大小为224*224的图像。
一种婴幼儿粪便采样图像分类处理方法，该方法基于一系统实现，所述系统包括有图像采集子系统和图像分类子系统，所述方法包括：步骤S1，利用所述图像采集子系统录入婴幼儿粪便图像；步骤S2，利用所述图像分类子系统对图像采集子系统录入的婴幼儿粪便图像进行分类训练，得出分类训练结果。
优选地，所述图像分类子系统包括有目标区域获取单元、图像裁剪单元、特征提取单元和分类训练单元，所述步骤S2的执行过程包括：步骤S20，利用所述目标区域获取单元获取婴幼儿粪便图像中的目标区域；步骤S21，利用所述图像裁剪单元在所述目标区域中裁剪出预设尺寸的图像；步骤S22，利用所述特征提取单元对裁剪得到的图像进行特征提取；步骤S23，所述分类训练单元根据所提取的图像特征，利用预设的分类模型对婴幼儿粪便图像进行分类训练，得出分类训练结果。
优选地，所述步骤S1中，所述图像采集子系统录入的婴幼儿粪便图像是像素大小为256*256的图像，所述步骤S21中，所述图像裁剪单元用于在所述目标区域中裁剪出像素大小为224*224的图像。
优选地，所述分类模型包括VGGNet16模型、ResNet50模型和GoogleNet 中的Inception V3模型，所述步骤S23中，分别利用所述VGGNet16模型、ResNet50模型和GoogleNet中的Inception V3模型进行分类训练。
优选地，所述步骤S23中的分类模型是卷积神经网络。
优选地，所述步骤S22中所提取的图像特征包括图像的纹理特征及颜色特征。
本发明公开的婴幼儿粪便采样图像分类处理系统中，首先获取婴幼儿粪便图像，从早期婴幼儿粪便图像中提取相关颜色、纹理等特征信息，再基于深度学习的方式对图像进行分类训练，进而得到分类训练结果，相比现有技术而言，本发明大大提高了图像分类识别的准确率，大大的提高了疾病筛查的效率，为早期婴幼儿疾病的筛查工作起到了较好的辅助作用。
附图说明
图1为本发明采样图像分类处理系统的组成框图；
图2为本发明采样图像分类处理方法的流程图。
具体实施方式
下面结合附图和实施例对本发明作更加详细的描述。
本发明公开了一种婴幼儿粪便采样图像分类处理系统，请参见图1，其包括有：
图像采集子系统1，用于录入婴幼儿粪便图像；
图像分类子系统2，用于对图像采集子系统1录入的婴幼儿粪便图像进行分类训练，得出分类训练结果。
上述系统中，首先获取婴幼儿粪便图像，从早期婴幼儿粪便图像中提取相关颜色、纹理等特征信息，再基于深度学习的方式对图像进行分类训练，进而 得到分类训练结果，相比现有技术而言，本发明大大提高了图像分类识别的准确率，大大的提高了疾病筛查的效率，为早期婴幼儿疾病的筛查工作起到了较好的辅助作用。
本实施例中，所述图像采集子系统1录入的图像包括用户通过图像摄录装置拍摄的婴幼儿粪便图像以及从医院收集的患儿粪便图像。
作为一种优选方式，所述图像分类子系统2包括：
目标区域获取单元20，用于获取婴幼儿粪便图像中的目标区域；
图像裁剪单元21，用于在所述目标区域中裁剪出预设尺寸的图像；
特征提取单元22，用于对裁剪得到的图像进行特征提取；
分类训练单元23，用于根据所提取的图像特征，利用预设的分类模型对婴幼儿粪便图像进行分类训练，得出分类训练结果。
进一步地，所述图像采集子系统1录入的婴幼儿粪便图像是像素大小为256*256的图像，所述图像裁剪单元21用于在所述目标区域中裁剪出像素大小为224*224的图像。
为了更好地描述本发明的技术方案，本发明还公开了一种婴幼儿粪便采样图像分类处理方法，结合图1和图2所示，该方法基于一系统实现，所述系统包括有图像采集子系统1和图像分类子系统2，所述方法包括：
步骤S1，利用所述图像采集子系统1录入婴幼儿粪便图像；
步骤S2，利用所述图像分类子系统2对图像采集子系统1录入的婴幼儿粪便图像进行分类训练，得出分类训练结果。
上述方法中，所述图像分类子系统2包括有目标区域获取单元20、图像裁剪单元21、特征提取单元22和分类训练单元23，所述步骤S2的执行过程包括：
步骤S20，利用所述目标区域获取单元20获取婴幼儿粪便图像中的目标区域；
步骤S21，利用所述图像裁剪单元21在所述目标区域中裁剪出预设尺寸的图像；
步骤S22，利用所述特征提取单元22对裁剪得到的图像进行特征提取；
步骤S23，所述分类训练单元23根据所提取的图像特征，利用预设的分类模型对婴幼儿粪便图像进行分类训练，得出分类训练结果。
进一步地，所述步骤S1中，所述图像采集子系统1录入的婴幼儿粪便图像是像素大小为256*256的图像，所述步骤S21中，所述图像裁剪单元21用于在所述目标区域中裁剪出像素大小为224*224的图像。
上述过程中，在进行训练时会出现过拟合的现象，为了减少过拟合对实验结果造成的影响，为此对数据进行数据增强，数据增强的原理是在原有的数据基础上通过一些技术手段来产生新的数据，由于粪便图像数据特点，在进行数据增强之前首选对数据进行感兴趣区域获取。在随后进行的数据增强中使用的是随机裁剪的方法，裁剪的结果是将输入大小为256*256的粪便图像随机裁剪成大小为224*224的图像。将大量婴幼儿粪便数据分布在虚拟云中，实现图像数据的高效安全存储和高速动态读写。随后对获得的粪便图像数据进行特征提取。
作为一种优选方式，所述分类模型包括VGGNet16模型、ResNet50模型和GoogleNet中的Inception V3模型，所述步骤S23中，分别利用所述VGGNet16模型、ResNet50模型和GoogleNet中的Inception V3模型进行分类训练。进一步地，训练过程中对模型的额参数进行微调，以获得较高的分类准确率，在对这三个分类模型进行参数微调之后，其训练过程的准确率可以达到 95％以上，在最后的测试结果中，其准确率也可达到90％以上。
作为一种替换方式，所述步骤S23中的分类模型是卷积神经网络。即利用卷积神经网络来对数据进行训练、测试，进而得到分类的结果。此外也可以采用机器学习二分类方法，如：支持向量机、K近邻算法来对粪便图像数据进行分类。
作为一种优选方式，所述步骤S22中所提取的图像特征包括图像的纹理特征及颜色特征。
本发明公开的婴幼儿粪便采样图像分类处理系统及方法，其可以应用在基于婴幼儿粪便图像数据进行分析处理的多种疾病筛查活动中，下面以婴幼儿早期胆道闭锁疾病筛查为例，对本发明的技术方案进行举例说明：
实施例一
本实施例中婴幼儿粪便采样图像分类处理系统的处理过程如下：首先利用所述图像采集子系统1录入婴幼儿粪便图像，从早期婴幼儿粪便图像中提取相关颜色、纹理等特征信息，再利用所述图像分类子系统2对图像采集子系统1录入的婴幼儿粪便图像进行分类训练，进而得到分类训练结果，上述处理过程有效提高了图像分类识别的准确率，同时可提高早期胆道闭锁的筛查效率，为早期婴幼儿胆道闭锁疾病的筛查工作起到了较好的辅助作用。
进一步地，所述图像采集子系统1录入的图像包括用户通过图像摄录装置拍摄的婴幼儿粪便图像以及从医院收集的胆道闭锁患儿粪便图像。
在具体的采集过程中，实验数据收集是经过医生或者病人家属对婴幼儿粪便图像进行拍照，随后再由医生对粪便图像进行标定(正常、异常)后得到数据。在数据标定的过程中有两位医生共同对数据进行标定，当两位医生对同一张粪便图像标定的结果一致时，才可区分这张图片为哪一类。当两名医生对同 一张粪便图像标定结果不一致的时候，我们就把这张图片忽略，不纳入数据集当中。
具体过程包括：建立大样本正常小孩和胆道闭锁患儿粪便图片数据库，不同手机型号和采集光照条件不同，对图像分析有着重要影响，而且日益增多的图像数据，数据存储和计算也都是很大挑战。通过对图像增强算法，研究基于云计算的医学影像信息存储分析平台，将大量婴幼儿粪便数据和诊断信息分布在虚拟云中，实现图像数据的高效安全存储和高速动态读写；研究基于云平台的数据并行计算，实现特征学习算法和学习分类算法的并行化，开发一套完整移动应用客户终端，实现信息共享，患者家属将数据上传到“云”存储，得到预判结果，患者也可以通过云平台进行专家会诊服务。
所述图像分类子系统2中，采用自动识别异常粪便图片的智能识别技术，通过计算机自动婴幼儿粪便图像量化诊断指标，更加准确、安全的辅助医生完成对疾病的诊断，研究利用深度学习的特征学习方法对图像特征的提取，研究利用弱监督的方法构建分类器对图像的分类。评价网络智能识别技术在胆道闭锁早期诊断中的价值，通过前期采集大样本临床案例，开发一套移动应用客户端，实际应用过程中，可联合多家治疗胆道闭锁的医学中心及多家妇幼保健中心共同参与项目对数据样本的采集，可以连续动态获取病变发展情况，并利用客户端对病例进行随访跟踪，最终评价和验证智能识别算法的可靠性、准确性。
本发明公开的婴幼儿粪便采样图像分类处理系统及方法，利用深度学习的方法对早期胆道闭锁等疾病进行筛查，能够充分的提取到粪便图像的特征，避免了使用粪便比色卡过程中出现的人为的错误因为的影响。同时，本发明可将深度学习算法集成到手机APP软件中，通过手机拍照自动识别正常或者异常的粪便图像，大大的提高了如早期胆道闭锁等疾病的筛查效率。更加准确、安 全的辅助医生完成对疾病的诊断。相比先由技术而言，本发明使用深度学习的方法进行筛查，先提取粪便图像特征再进行训练测试，大大的提高了分类识别的准确率。
应当说明的是，本发明提出的婴幼儿粪便采样图像分类处理系统及方法，其实际属于婴幼儿疾病筛查过程中用于实现数据统计与处理的计算机系统和计算机方法，并非某种特定疾病的治疗和筛查手段，同时，本发明对具体的应用场景不作限制，即无论将本发明作为任何一种婴幼儿疾病筛查的辅助计算机解决方案，均应当属于本发明的保护范围。

Claims (10)

  1. 一种婴幼儿粪便采样图像分类处理系统，其特征在于，包括有：
    图像采集子系统(1)，用于录入婴幼儿粪便图像；
    图像分类子系统(2)，用于对图像采集子系统(1)录入的婴幼儿粪便图像进行分类训练，得出分类训练结果。
  2. 如权利要求1所述的婴幼儿粪便采样图像分类处理系统，其特征在于，所述图像采集子系统(1)录入的图像包括用户通过图像摄录装置拍摄的婴幼儿粪便图像以及从医院收集的患儿粪便图像。
  3. 如权利要求1所述的婴幼儿粪便采样图像分类处理系统，其特征在于，所述图像分类子系统(2)包括：
    目标区域获取单元(20)，用于获取婴幼儿粪便图像中的目标区域；
    图像裁剪单元(21)，用于在所述目标区域中裁剪出预设尺寸的图像；
    特征提取单元(22)，用于对裁剪得到的图像进行特征提取；
    分类训练单元(23)，用于根据所提取的图像特征，利用预设的分类模型对婴幼儿粪便图像进行分类训练，得出分类训练结果。
  4. 如权利要求3所述的婴幼儿粪便采样图像分类处理系统，其特征在于，所述图像采集子系统(1)录入的婴幼儿粪便图像是像素大小为256*256的图像，所述图像裁剪单元(21)用于在所述目标区域中裁剪出像素大小为224*224的图像。
  5. 一种婴幼儿粪便采样图像分类处理方法，其特征在于，该方法基于一系统实现，所述系统包括有图像采集子系统(1)和图像分类子系统(2)，所述方法包括：
    步骤S1，利用所述图像采集子系统(1)录入婴幼儿粪便图像；
    步骤S2，利用所述图像分类子系统(2)对图像采集子系统(1)录入的婴幼儿粪便图像进行分类训练，得出分类训练结果。
  6. 如权利要求5所述的婴幼儿粪便采样图像分类处理方法，其特征在于，所述图像分类子系统(2)包括有目标区域获取单元(20)、图像裁剪单元(21)、特征提取单元(22)和分类训练单元(23)，所述步骤S2的执行过程包括：
    步骤S20，利用所述目标区域获取单元(20)获取婴幼儿粪便图像中的目标区域；
    步骤S21，利用所述图像裁剪单元(21)在所述目标区域中裁剪出预设尺寸的图像；
    步骤S22，利用所述特征提取单元(22)对裁剪得到的图像进行特征提取；
    步骤S23，所述分类训练单元(23)根据所提取的图像特征，利用预设的分类模型对婴幼儿粪便图像进行分类训练，得出分类训练结果。
  7. 如权利要求6所述的婴幼儿粪便采样图像分类处理方法，其特征在于，所述步骤S1中，所述图像采集子系统(1)录入的婴幼儿粪便图像是像素大小为256*256的图像，所述步骤S21中，所述图像裁剪单元(21)用于在所述目标区域中裁剪出像素大小为224*224的图像。
  8. 如权利要求6所述的婴幼儿粪便采样图像分类处理方法，其特征在于，所述分类模型包括VGGNet16模型、ResNet50模型和GoogleNet中的Inception V3模型，所述步骤S23中，分别利用所述VGGNet16模型、ResNet50模型和GoogleNet中的Inception V3模型进行分类训练。
  9. 如权利要求6所述的婴幼儿粪便采样图像分类处理方法，其特征在于，所述步骤S23中的分类模型是卷积神经网络。
  10. 如权利要求6所述的婴幼儿粪便采样图像分类处理方法，其特征在于，所述步骤S22中所提取的图像特征包括图像的纹理特征及颜色特征。
PCT/CN2019/130562 2019-05-05 2019-12-31 一种婴幼儿粪便采样图像分类处理系统及方法 Ceased WO2020224282A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910366945.8 2019-05-05
CN201910366945.8A CN110222713A (zh) 2019-05-05 2019-05-05 一种婴幼儿粪便采样图像分类处理系统及方法

Publications (1)

Publication Number Publication Date
WO2020224282A1 true WO2020224282A1 (zh) 2020-11-12

Family

ID=67820311

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/130562 Ceased WO2020224282A1 (zh) 2019-05-05 2019-12-31 一种婴幼儿粪便采样图像分类处理系统及方法

Country Status (2)

Country Link
CN (1) CN110222713A (zh)
WO (1) WO2020224282A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023283252A1 (en) * 2021-07-06 2023-01-12 Toi Labs, Inc. System, method and apparatus to detect health abnormality in patients
US12089822B2 (en) 2017-04-07 2024-09-17 Toi Labs, Inc. Biomonitoring devices, methods, and systems for use in a bathroom setting
US12440193B2 (en) 2017-04-07 2025-10-14 Toi Labs, Inc. System, method and apparatus for forming machine learning sessions

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110222713A (zh) * 2019-05-05 2019-09-10 深圳先进技术研究院 一种婴幼儿粪便采样图像分类处理系统及方法
CN111183917B (zh) * 2019-11-14 2022-02-01 京东科技信息技术有限公司 一种动物异常监测、图像处理方法及装置
WO2021261096A1 (ja) * 2020-06-26 2021-12-30 パナソニック株式会社 出力装置、方法、プログラム、及び表示装置
CN112002415B (zh) * 2020-08-23 2024-03-01 吾征智能技术(北京)有限公司 一种基于人体粪便智能认知疾病系统
CN112183674B (zh) * 2020-11-06 2022-06-10 南昌航空大学 一种粪便宏观图像颜色和性状多任务识别方法及系统
CN113593696A (zh) * 2021-07-12 2021-11-02 金世柱 粪便自主筛选诊断系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120316421A1 (en) * 2009-07-07 2012-12-13 The Johns Hopkins University System and method for automated disease assessment in capsule endoscopy
CN106682633A (zh) * 2016-12-30 2017-05-17 四川沃文特生物技术有限公司 基于机器视觉的粪便镜检图像有形成分的分类识别方法
CN107066779A (zh) * 2016-10-18 2017-08-18 应继伟 基于生物样本颜色的自助身体参数检查方法及系统
CN108009567A (zh) * 2017-11-10 2018-05-08 电子科技大学 一种结合图像颜色及hog和svm的粪便性状的自动辨别方法
CN110222713A (zh) * 2019-05-05 2019-09-10 深圳先进技术研究院 一种婴幼儿粪便采样图像分类处理系统及方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120031976A1 (en) * 2010-08-06 2012-02-09 Yan Bao Approach of disposing waste
CN106651883B (zh) * 2016-12-30 2019-12-17 四川沃文特生物技术有限公司 基于机器视觉的粪便形态识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120316421A1 (en) * 2009-07-07 2012-12-13 The Johns Hopkins University System and method for automated disease assessment in capsule endoscopy
CN107066779A (zh) * 2016-10-18 2017-08-18 应继伟 基于生物样本颜色的自助身体参数检查方法及系统
CN106682633A (zh) * 2016-12-30 2017-05-17 四川沃文特生物技术有限公司 基于机器视觉的粪便镜检图像有形成分的分类识别方法
CN108009567A (zh) * 2017-11-10 2018-05-08 电子科技大学 一种结合图像颜色及hog和svm的粪便性状的自动辨别方法
CN110222713A (zh) * 2019-05-05 2019-09-10 深圳先进技术研究院 一种婴幼儿粪便采样图像分类处理系统及方法

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US12089822B2 (en) 2017-04-07 2024-09-17 Toi Labs, Inc. Biomonitoring devices, methods, and systems for use in a bathroom setting
US12440193B2 (en) 2017-04-07 2025-10-14 Toi Labs, Inc. System, method and apparatus for forming machine learning sessions
WO2023283252A1 (en) * 2021-07-06 2023-01-12 Toi Labs, Inc. System, method and apparatus to detect health abnormality in patients

Also Published As

Publication number Publication date
CN110222713A (zh) 2019-09-10

Similar Documents

Publication Publication Date Title
WO2020224282A1 (zh) 一种婴幼儿粪便采样图像分类处理系统及方法
Liu et al. Referable diabetic retinopathy identification from eye fundus images with weighted path for convolutional neural network
Benčević et al. Understanding skin color bias in deep learning-based skin lesion segmentation
CN107423571B (zh) 基于眼底图像的糖尿病视网膜病变识别系统
CN107358606B (zh) 用于识别糖尿病视网膜病变的人工神经网络装置及系统装置
WO2019196268A1 (zh) 基于深度学习的糖尿病视网膜图像分类方法及系统
CN108464840A (zh) 一种乳腺肿块自动检测方法及系统
CN117116477B (zh) 基于随机森林和XGBoost的前列腺癌患病风险预测模型的构建方法及系统
Lorenzoni et al. Neonatal screening for congenital cytomegalovirus infection in preterm and small for gestational age infants
CN111785363A (zh) 一种基于ai引导的慢性病辅助诊断系统
CN111883256A (zh) 基于电子病历数据的肺结核患者预警系统及预警方法
Hatanaka et al. Improvement of automated detection method of hemorrhages in fundus images
Khan et al. RVD: a handheld device-based fundus video dataset for retinal vessel segmentation
CN114400087A (zh) 基于检验数据与知识图谱诊断布鲁氏菌病的方法
CN113539476A (zh) 基于人工智能的胃内窥活检拉曼图像辅助诊断方法和系统
CN115661101A (zh) 基于随机采样和深度学习的早产儿视网膜病变检测系统
CN110215232A (zh) 基于目标检测算法的冠状动脉血管内超声斑块分析方法
CN114913169B (zh) 一种新生儿坏死性小肠结肠炎筛查系统
Michelle et al. Covid-19 detection from X-rays with aid of convolutional neural network
CN116309461A (zh) 一种基于人工智能的心脏瓣口分割与面积计算方法及装置
Loja-Morocho et al. Intelligent system to provide support in the analysis of colposcopy images based on artificial vision and deep learning: A first approach for rural environments in Ecuador
Lim et al. A review of machine learning techniques for applied eye fundus and tongue digital image processing with diabetes management system
JP7845702B2 (ja) 両眼の偏向を利用した急性大血管閉塞検査装置および方法
Ji et al. A semantic segmentation-based automatic pterygium assessment and grading system
Rocha et al. Segmentation of COVID-19 lesions in CT Images

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19927784

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19927784

Country of ref document: EP

Kind code of ref document: A1