iMedBot A Web -based Intelligent Agent for Healthcare Related Prediction and Deep Learning Chuhan Xu Xia Jiang Abstract

2025-04-24 0 0 424.66KB 6 页 10玖币

侵权投诉

iMedBot: A Web-based Intelligent Agent for Healthcare Related Prediction and Deep Learning

Chuhan Xu, Xia Jiang

Abstract:

Background: Breast cancer is a multifactorial disease, genetic and environmental factors will affect its

incidence probability. Breast cancer metastasis is one of the main cause of breast cancer related deaths

reported by the American Cancer Society (ACS). Method: the iMedBot is a web application that we

developed using the python Flask web framework and deployed on Amazon Web Services. It contains a

frontend and a backend. The backend is supported by a python program we developed using the python

Keras and scikit-learn packages, which can be used to learn deep feedforward neural network (DFNN)

models. Result: the iMedBot can provide two main services: 1. it can predict 5-, 10-, or 15-year breast

cancer metastasis based on a set of clinical information provided by a user. The prediction is done by using

a set of DFNN models that were pretrained, and 2. It can train DFNN models for a user using user-provided

dataset. The model trained will be evaluated using AUC and both the AUC value and the AUC ROC curve

will be provided. Conclusion: The iMedBot web application provides a user-friendly interface for user-

agent interaction in conducting personalized prediction and model training. It is an initial attempt to convert

results of deep learning research into an online tool that may stir further research interests in this direction.

Keywords: Deep learning, Breast Cancer, Web application, Model training.

1. Introduction

This paper focuses on an introduction of the iMedBot – A web-based Intelligent Agent, which was initially

developed as an user friendly and interactive online agent for predicting n-year breast cancer metastasis.

The use of the current version of iMedBot is limited to the research community for the purpose of boosting

the deployment and dissemination

of research results concerning deep

learning, machine learning, and

other methods of artificial

intelligence (AI), further stirring

research interests in AI in

medicine, and setting an example

of a web-based intelligent agent

that can assist medical activities

such as prognosis and decision

support. As shown in Figure 1, the

iMedBot is a full-stack web

application that consists of both a

front end a back end. The back end

contains the model training service

module, the model prediction

service model, the serialized model object, and the KerasClassifier module. The later two S3 isare the

supporting hardware components developed by using the resources provided by the Amazon web services

(AWS). The current core services provided by the back end include 1) Model training service; and 2)

personalized prediction of 5-, 10-, and 15-year breast cancer metastasis. The front end of the iMedBot

consists of the client, the domain name server (DNS), and a load Balancer. The route53, ELB, EC2later

two are also developed usiAWS services resources to support the main functions of the front end. The client

module provides the agent-user interface, which supports the agent-user interaction by a set of

“conversation” windows. The main window hosts the sequence of dialogues once the “conversation”

between the again and user begins.

Figure 1. The system structure of the iMedBot.

Client Load Balancer

(AWS ELB)

DNS

(AWS route53)

Web Server

(AWS EC2)

Model Training

Service

Model Prediction

Service

KerasClassifier Module

(AWS S3)

Serialized Model Object (AWS

S3)

The System Structure of iMedBot

FrontEnd BackEnd

2. The Backend Services

The model training service is provided by our python deep feedforward neural network (DFNN) programs

[1], [2], which were developed by using the Keras python package [3]. We included this service in the

iMedBot because we assume there are scenarios in which the users prefer training prediction models using

their own data. When a model training service call is initialized in the front end and passed to the back end,

the dataset provided by the user will be split in stratified manner: 80% of the data will be used to train

models following the 5-fold cross validation strategies and 20% of the data will be used as the validation

dataset. The validation AUC is one of the output of the model training service, which will be displayed by

the [4].which will be generated by testing the best output model of a grid search using the validation dataset.

Grid search is a systematic way of conduct hyperparameter tuning to identify the best performing models

in deep learning [3]. Our DFNN models have 13 hyperparameters. We call one particular value assignment

of the set of hyperparameters of our DFNN models a hyperparameter setting. So each DFNN model has a

unique hyperparameter setting. Before a grid search, each of the hyperparameters is given a range of values

the hyperparameter can take. Grid search will train all DFNN models corresponding to all possible

hyperparameter settings determined by the preselected ranges of values for the hyperparameters. The results

returned by the model training service include the validation AUC, the ROC curve plot [4], and the best

prediction model found by grid search.

The prediction service is provided by a set of deep feedforward neural network (DFNN) models

that were pretrained with our DFNN programs [3] by Grid search strategy.[3] Grid search will train all

DFNN models corresponding to all possible hyperparameter settings determined by the preselected ranges

of values for the hyperparameters. The same procedures used in model training service were used for

pretraining these models. The difference between these two kinds of services is that the model training

service is meant to train models using user-provided datasets, while the pretrained DFNN models were

trained using the LSM 5-, 10-, 15-year datasets that are publicly available [3], [5]. The pretrained DFNN

models are the best performing models selected from a large set of models trained via grid search. A detailed

description of the DFNN model training, grid search, and model evaluation is provided in [3], [6]. The

pretrained DFNN models contain a set of predictors [3], which are the clinical features defined by the

datasets from which these models were trained; These models can be used to conduct personalized

prediction once the patient-specific values of the predictors are received.

In addition to the two core services, the back end contains other python tools that we developed for

tasks such as processing input data, analyzing results, and evaluating the prediction performance of models.

These tools provide some of the assistant services such as generating the ROC curves based on the 5-fold

cross validation to compare the prediction performance of different models [3]. The iMedBot is currently

hosted at the AWS (Amazon Web Services), and therefore there are other supporting system components

in the backend, which are provided by AWS S3.

3. The Front End

The front end of the iMedBot has a main user friendly agent-user interaction window, which hosts the

sequence of dialogues once the “conversion” between the again and user begins. The possible dialogues are

designed based on the current two core back end services that the iMedBot can provide: the prediction and

model training. For example, to provide the personalized prediction service, the iMedBot will initialize a

sequence of dialogues to go through with the user the set of the predictors of the back end DFNN models

one after another. For each of the predictors, the iMedBot will provide to the user a list of all possible values

of the predictor, from which ana user can select his/her patient specific value with an action as simple as

clicking a button. Once the iMedBot receives the user input/responses, it will communicate with the proper

back end components, pass over the user input collected through the “conversation” to the best model object

that we deployed in our S3 bucket in the backend, and then receive and display the results from the back

end. Figure 2, 3 show two examples of these agent-user interaction dialogues. The first example shows the

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

iMedBot:AWeb-basedIntelligentAgentforHealthcareRelatedPredictionandDeepLearningChuhanXu,XiaJiangAbstract:Background:Breastcancerisamultifactorialdisease,geneticandenvironmentalfactorswillaffectitsincidenceprobability.BreastcancermetastasisisoneofthemaincauseofbreastcancerrelateddeathsreportedbytheAm...

展开>> 收起<<

iMedBot A Web -based Intelligent Agent for Healthcare Related Prediction and Deep Learning Chuhan Xu Xia Jiang Abstract.pdf

共6页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

iMedBot A Web -based Intelligent Agent for Healthcare Related Prediction and Deep Learning Chuhan Xu Xia Jiang Abstract

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: