Study On Automated Diagnostic Models By Mining Association Rules From SPECT Textual Diagnostic Reports

Medical imaging is a technical means of presenting the structure and function of the body.Medical imaging mainly includes traditional structural imaging,functional imaging,and hybrid imaging.SPECT functional imaging requires the injection of99 m Tc-MDP radiopharmaceuticals in the body.After a certain period of time,there will be drug residues in the body,which then shows the distribution of systemic radiation.Due to the confidentiality of the accumulated massive medical data,which cannot be shared with the outside world in real time,the external personnel need high cost to obtain it,which makes the development of this part of data lag.In addition,medical staff do not have the extra ability and energy to carry out information mining on this part of data.The current diagnosis of nuclear medicine diseases mainly relies on manual image reading,and the academic community is developing various methods to overcome the inefficiency of manual image reading and diagnosis.Starting from the current application status of nuclear medicine diagnosis,such as low efficiency and high false alarms,the bone scan nuclear medicine text is used as the research object,and machine learning is used in natural language processing.The research content is as follows:(1)In order to explore the relationship between diseases and their manifestations,a nuclear medicine text mining method based on Apriori algorithm is researched and proposed.First,in view of the information redundancy,missing data,and inconsistent expressions that may be contained in nuclear medicine diagnostic texts,a preprocessing and unified coding method for SPECT nuclear medicine diagnostic texts is proposed;then,the traditional Apriori algorithm is applied to propose the relationship between the lesion and the representation.Association mining algorithm;finally,using a set of real SPECT nuclear medicine diagnosis text data from the nuclear medicine department of the top three hospitals,the method proposed in this paper is verified,and the results show that the method proposed in this paper objectively extracts the association between the disease and its characterization.The average value of the obtained objective evaluation index reaches 90%.(2)Aiming at the low efficiency and high false alarms of nuclear medicine diagnosis,a nuclear medicine text diagnosis model based on deep learning is researched and proposed.First,use different methods to preprocess the data in the text.Then,using the traditional machine learning algorithm random forest and SVM,the genetic algorithm optimization classification method of lesions and their representations is proposed.Secondly,using the classic deep learning algorithm Text CNN,a classification method based on CNN is proposed.Finally,using a set of real nuclear medicine text data to verify the method proposed in this paper,the experimental results show that the method proposed in this paper objectively classifies several types of diseases,and the objective evaluation index obtained reaches 94%.(3)In order to assist clinicians in diagnosing diseases,a nuclear medicine auxiliary diagnosis system is designed and implemented,which can realize frequent item sets,correlation extraction of diseases and their manifestations,and the acquisition of auxiliary diagnosis evaluation results.Tests and experiments show that this system can achieve the goal of text research.From the perspective of data mining,machine learning methods are used to classify and predict nuclear medicine text data,thereby establishing a diagnostic model,achieving the effect of assisting doctors in diagnosis,and reducing the pressure on clinicians to a certain extent.