Research And Implementation Of Intelligent Recongnition System For Medical Laboratory Data

In recent years,with the progress and development of society,the reform and development of the medical field has become a social hotspot.As a part of the national strategy of building a "healthy China",the telemedicine service of graded diagnosis and treatment is gradually advancing,and the use of artificial intelligence technology to promote the reform and development of the medical field has also gradually attracted attention.In the graded diagnosis and treatment program,the establishment of the medical records of medical patients in multi-level hospitals is the basic work.Generally speaking,the patient's medical records mainly include:laboratory records,medical records,laboratory images,etc.However,most grassroots hospitals currently do not have a complete electronic medical record database.In the case of graded diagnosis and treatment,most of the patients take the printed test list into a photo and send it to the remote doctor in the form of a picture.Such image data is unstructured data.From the perspective of establishing a patient's medical recor-d file,how to extract and structure the information in the picture of laboratory sheet is an urgent need.This research project will use the rapidly developing deep learning technology to establish a complete system for identifying and extracting information from the picture of laboratory sheet,and implement an end-to-end system from image input to structured data output.In terms of algorithm,this paper designs and implements the text region detection model and string recognition model,and mainly studies and improves the string recognition model,include the Chinese string recognition model based on convolution neural network and the non-Chinese string recognition model based on sequence model.The center loss in the field of face recognition and the focal loss in the field of target detection are introduced innovatively,and a new example for the text recognition task is constructed,which shows that it also has great potential in the field of text recognition.At the same time,from the perspective of model lightweighting,this paper design a group convolution module to replace the traditional convolution operation,thus achieving a reduction in model parameters and computational complexity.Moreover,this paper draws on the idea of metric learning and model uncertainty,proposes an uncertainty metric model.The model gives the prediction uncertainty while giving the prediction results and satisfies the need for the uncertainty assessment of the model identification results in practical applications.In engineering,this paper constructs a complete online test list identification system based on Django Web framework,in combination with the text region detection model,string recognition model and uncertainty measurement model.The system structure includes two parts:an offline model training module and online system module,and this system can be used to complete the identification of the test list.