Search

BACKGROUND Pre-training large-scale neural language models on raw texts has made a significant contribution to improving transfer learning in natural language processing. With the introduction of transformer-based language models, such as bidirectional encoder representations from transformers (BERT), the performance of information extraction from free text has significantly improved for both the general and medical domains; however, it is difficult to train specific BERT models that perform well for domains in which there are few publicly available databases of high quality and large size. OBJECTIVE We hypothesize that this problem can be addressed by over-sampling a domain-specific corpus and using it for pre-training with a larger corpus in a balanced manner. In this study, we verify our hypothesis by developing pre-training models using our method and evaluating their performance. METHODS Our proposed method was based on simultaneous pre-training after over-sampling. We conducted three experiments in which we generated (1) English biomedical BERT from a small biomedical corpus, (2) Japanese medical BERT from a small medical corpus, and (3) enhanced biomedical BERT pre-trained from complete PubMed abstracts in a balanced manner and compared their performance with the conventional models. RESULTS We first confirmed that our English BERT pre-trained using both general and small medical-domain corpora performed sufficiently well for practical use in the biomedical language understanding evaluation (BLUE) benchmark. Moreover, our proposed method was more effective than conventional methods for each different biomedical corpus size with the same corpus size for the general domain. Next, our Japanese medical BERT outperformed the other BERT models built using a conventional method concerning the medical document classification task. It demonstrated the same trend as in the first experiment in English. Lastly, our enhanced biomedical BERT model, in which clinical notes were not used during pre-training, achieved both clinical and biomedical scores on the BLUE benchmark that were 0.3 points above those of the model trained without our proposed method. CONCLUSIONS Well-balanced pre-training by over-sampling instances derived from a corpus appropriate for the target task allowed us to construct a high-performance BERT model.

Background The role of patients in medical research is changing, as more emphasis is being placed on patient involvement, and patient reported outcomes are increasingly contributing to clinical decision-making. Information and communication technology provides new opportunities for patients to actively become involved in research. These trends are particularly noticeable in Europe and the US, but less obvious in Japan. The aim of this study was to investigate the practice of active involvement of patients in medical research in Japan by utilizing a digital platform, and to analyze the outcomes to clarify what specific approaches could be put into practice. Methods We developed the RUDY JAPAN system, an ongoing rare disease medical research platform, in collaboration with the Rare and Undiagnosed Diseases Study (RUDY) project in the UK. After 2 years of preparation, RUDY JAPAN was launched in December 2017. Skeletal muscle channelopathies were initially selected as target diseases, and hereditary angioedema was subsequently added. Several approaches for active patient involvement were designed through patient-researcher collaboration, namely the Steering Committee, questionnaire development, dynamic consent, and other communication strategies. We analyzed our practices and experiences focusing on how each approach affected and contributed to the research project. Results RUDY JAPAN has successfully involved patients in this research project in various ways. While not a part of the initial decision-making phase to launch the project, patients have increasingly been involved since then. A high level of patient involvement was achieved through the Steering Committee, a governance body that has made a major contribution to RUDY JAPAN, and the process of the questionnaire development. The creation of the Patient Network Forum, website and newsletter cultivated dialogue between patients and researchers. The registry itself allowed patient participation through data input and control of data usage through dynamic consent. Conclusions We believe the initial outcomes demonstrate the feasibility and utility of active patient involvement in Japan. The collaboration realized through RUDY JAPAN was enabled by digital technologies. It allowed busy patients and researchers to find the space to meet and work together for the Steering Committee, questionnaire development and various communication activities. While the practice of active patient involvement in Japan is still in its early stages, this research confirms its viability if the right conditions are in place. (331 words).

Intracranial artery stenosis is the predominant etiology of ischemic stroke in the Asian population. Furthermore, the presence of theUsing a prospective database, we identified adult patients with intracranial artery stenosis who underwent periodic MRI examinations for5 years. We evaluated stenosis progression using a validated visual grading system. We excluded patients diagnosed with moyamoya disease at the time of initial MRI. Genotyping ofAmong 52 eligible patients, 22 (42%) were carriers of theOur findings indicated that the

BACKGROUND Medicines may cause various adverse reactions. An enormous amount of money and effort is spent investigating adverse drug events (ADEs) in clinical trials and postmarketing surveillance. Real-world data from multiple electronic medical records (EMRs) can make it easy to understand the ADEs that occur in actual patients. OBJECTIVE In this study, we generated a patient medication history database from physician orders recorded in EMRs, which allowed the period of medication to be clearly identified. METHODS We developed a method for detecting ADEs based on the chronological relationship between the presence of an adverse event and the medication period. To verify our method, we detected ADEs with alanine aminotransferase elevation in patients receiving aspirin, clopidogrel, and ticlopidine. The accuracy of the detection was evaluated with a chart review and by comparison with the Roussel Uclaf Causality Assessment Method (RUCAM), which is a standard method for detecting drug-induced liver injury. RESULTS The calculated rates of ADE with ALT elevation in patients receiving aspirin, clopidogrel, and ticlopidine were 3.33% (868/26,059 patients), 3.70% (188/5076 patients), and 5.69% (226/3974 patients), respectively, which were in line with the rates of previous reports. We reviewed the medical records of the patients in whom ADEs were detected. Our method accurately predicted ADEs in 90% (27/30patients) treated with aspirin, 100% (9/9 patients) treated with clopidogrel, and 100% (4/4 patients) treated with ticlopidine. Only 3 ADEs that were detected by the RUCAM were not detected by our method. CONCLUSIONS These findings demonstrate that the present method is effective for detecting ADEs based on EMR data.

BackgroundMild cognitive impairment (MCI) is a high-risk condition for conversion to dementias, including Alzheimer's disease (AD) dementia. However, individuals with MCI show heterogeneity in patterns of pathology, and MCI does not always convert to AD dementia. Detailed subtyping of MCI and accurate prediction of the patients in whom MCI will convert to AD dementia may support new trial designs and enable evaluation of the efficacy of drugs within small numbers of patients during clinical trials. MethodsWe constructed a decision tree model by the heterogeneous mixture learning (HML) method, integrating cerebrospinal fluid (CSF) biomarker data, structural MRI data, APOE genotype data, and a recorded age at examination. The decision tree model was applied to predict conversion to AD dementia and to identify subtypes of MCI. After the test performances of HML models were assessed, MCI subjects were classified into some subtypes based on a decision tree. Then, we characterized each MCI subtype in terms of the degree of CSF biomarker abnormalities and brain atrophy, declines of cognitive functions, and gene expression alterations derived from peripheral blood samples.ResultsWe identified five subtypes of MCI using the HML approach and categorized them into three groups: those similar to CN subjects with low conversion rates; those with intermediate conversion rates; and those similar to patients with AD with high conversion rates. Furthermore, the subtypes with intermediate conversion rates were separated into the subtype with CSF biomarker abnormalities or the subtype with brain atrophy. The results from the CSF inflammation marker and gene expression analysis suggested the occurrence of aberrant inflammatory immune responses in the CSF and blood of the subjects in the subtypes with CSF biomarker abnormalities. ConclusionThe subtypes that were identified in this study exhibited varying conversion rates to AD as well as differing levels of biological features. Focusing on specific subtypes in which conversion to AD can be predicted with the most accuracy could enable more efficient clinical trials to be conducted.

Background and objective In this study, we tried to create a machine-learning method that detects disease lesions from chest X-ray (CXR) images using a data set annotated with extracted CXR reports information. We set the nodule as the target disease lesion. Manually annotating nodules is costly in terms of time. Therefore, we used the report information to automatically produce training data for the object detection task. Methods First, we use semantic segmentation model PSP-Net to recognize lung fields described in the CXR reports. Next, a classification model ResNeSt-50 is used to discriminate the nodule in segmented right and left field. It also can provide attention map by Grad-Cam. If the attention region corresponds to the location of the nodule in the CXR reports, an attention bounding box is generated. Finally, object detection model Faster-RCNN was performed using generated attention bounding box. The bounding boxes predicted by Faster-RCNN were filtered to satisfy the location extracted from CXR reports. Results For lung field segmentation, a mean intersection of union of 0.889 was achieved in our best model. 15,156 chest radiographs are used for classification. The area under the receiver operating characteristics curve was 0.843 and 0.852 for the left and right lung, respectively. The detection precision of the generated attention bounding box was 0.341 to 0.531 depending on the binary setting for attention map. Through object detection process, the detection precisions of the bounding boxes were improved to 0.567 to 0.800. Conclusion We successfully generated bounding boxes with nodule on CXR images based on the positional information of the diseases extracted from the CXR reports. Our method has the potential to provide bounding boxes for various lung lesions which can reduce the annotation burden for specialists. Short abstract Machine learning for computer aided image diagnosis requires annotation of images, but manual annotation is time-consuming for medical doctor. In this study, we tried to create a machine-learning method that creates bounding boxes with disease lesions on chest X-ray (CXR) images using the positional information extracted from CXR reports. We set the nodule as the target lesion. First, we use PSP-Net to segment the lung field according to the CXR reports. Next, a classification model ResNeSt-50 was used to discriminate the nodule in segmented lung field. We also created an attention map using the Grad-Cam algorithm. If the area of attention matched the area annotated by the CXR report, the coordinate of the bounding box was considered as a possible nodule area. Finally, we used the attention information obtained from the nodule classification model and let the object detection model trained by all of the generated bounding boxes. Through object detection model, the precision of the bounding boxes to detect nodule is improved.

The acquisition of medical images from multiple medial institutions has become important for high-quality clinical studies. In recent years, electronic data submission has enabled the transmission of image data to independent institutions more quickly and easily than before. However, the selection, anonymization, and transmission of medical images still require human resources in the form of clinical research collaborators. In this study, we developed an image collection system that works with the electronic data capture (EDC) system. In this image collection system, medical images are selected based on EDC input information, patient ID is anonymized to a subject ID issued by the EDC, and the selected anonymized images are transferred to the research institute without human intervention. In the research institute, clinical information registered by the EDC and clinical images collected by the image collection system are managed by the same subject ID and can be used for clinical studies. In October 2019, our image collection system was introduced to 13 medical institutions and has now begun collecting medical images from the in-hospital picture archiving and communication system (PACS) of those institutions.

Background Medicines may cause various adverse reactions. An enormous amount of money and effort is spent investigating adverse drug events (ADEs) in clinical trials and postmarketing surveillance. Real-world data from multiple electronic medical records (EMRs) can make it easy to understand the ADEs that occur in actual patients. Objective In this study, we generated a patient medication history database from physician orders recorded in EMRs, which allowed the period of medication to be clearly identified. Methods We developed a method for detecting ADEs based on the chronological relationship between the presence of an adverse event and the medication period. To verify our method, we detected ADEs with alanine aminotransferase elevation in patients receiving aspirin, clopidogrel, and ticlopidine. The accuracy of the detection was evaluated with a chart review and by comparison with the Roussel Uclaf Causality Assessment Method (RUCAM), which is a standard method for detecting drug-induced liver injury. Results The calculated rates of ADE with ALT elevation in patients receiving aspirin, clopidogrel, and ticlopidine were 3.33% (868/26,059 patients), 3.70% (188/5076 patients), and 5.69% (226/3974 patients), respectively, which were in line with the rates of previous reports. We reviewed the medical records of the patients in whom ADEs were detected. Our method accurately predicted ADEs in 90% (27/30patients) treated with aspirin, 100% (9/9 patients) treated with clopidogrel, and 100% (4/4 patients) treated with ticlopidine. Only 3 ADEs that were detected by the RUCAM were not detected by our method. Conclusions These findings demonstrate that the present method is effective for detecting ADEs based on EMR data.

Background Electronic medical records (EMRs) are widely used, but in many cases, they are used within a network physically separated from the Internet. Multicenter clinical studies use Internet-connected electronic data capture (EDC) systems to collect data, where data entered into the EMR are manually transcribed into the EDC system. In addition, medical images for clinical research are also collected manually. Variations in EMRs and differing data structures among vendors hamper the use of data for clinical research. Methods We solved this problem by developing a network infrastructure for clinical research between Osaka University Hospital and affiliated hospitals in the Osaka area and introducing a clinical data collection system (CDCS). In each hospital's EMR network, we implemented a CRF reporter that accumulated data for clinical research using a template and then sent the data to a management server in the Osaka University Hospital Data Center. To organize the patient profile data and clinical laboratory data stored in each EMR for use in clinical research, the data are retrieved from the template by an interface module developed by each vendor, according to our common data output interface specification. The data entered into the CRF reporter template for clinical research are also recorded in the EMR progress notes and sent to the data management server. This network infrastructure can also be used as a medical image collection system that automatically collects images for research from PACS at each hospital. These systems are managed under common subject numbers issued by the CDCS. Results A network infrastructure was established among 19 hospitals, and a CRF reporter was incorporated into the EMR. A medical image transfer system was introduced in 13 hospitals. Since 2013, 28 clinical studies have been conducted using this system, and data for 9,987 cases have been collected as of December 31, 2020. Conclusion Incorporating a CRF reporter with medical image transfer system into the EMR has proven useful for collecting research data.

Our hospital stores all clinical records as Portable Document Formats (PDFs). These PDFs are delivered by each system with a document profile XML file. Using this interface, the items thought to be important for clinical studies are described in the document profile XML and delivered to the data warehouse (DWH). In case clinical data not stored in the DWH are needed, we extract the data from PDF documents. Even from scanned PDFs, the data can be extracted with high accuracy.

Early diagnosis and treatment of pancreatic cancer is challenging. We attempted to find diagnostic rules for pancreatic cancer from laboratory data in the Osaka University Hospital's data warehouse using Bayesian estimation. We calculated the pretest odds based on the number of laboratory tests and the cutoff value at which the diagnostic accuracy is over 20%. By this method, we identified diagnostic rules of 6 types for one item and 79 types for 2 items. Pancreatic cancer is difficult to detect from only general laboratory tests. However, this method may be promising in early diagnosis.

Electronic health record (EHR) systems are necessary for the sharing of medical information between care delivery organizations (CDOs). We developed a document-based EHR system in which all of the PDF documents that are stored in our electronic medical record system can be disclosed to selected target CDOs. An access control list (ACL) file was designed based on the HL7 CDA header to manage the information that is disclosed.

Background: Electronic health record (EHR) systems are necessary for sharing medical information between care delivery organizations (CDOs). While many standardized data can be shared, it is still difficult to share nonstandardized clinical data. Furthermore, there remains the problem that the EHR system cannot finely control the disclosure status. We have been operating a documentbased electronic medical record system called the Document Archiving and Communication System (DACS) in which printed images of all medical records are stored in PDF format. Objectives: To develop a document-based EHR system that can disclose selected clinical documents in PDF format. Methods: With a document-based EHR system, any documents stored in the DACS can be opened to a designated CDO. These open documents can be selected by the service CDO staff. In order to manage the disclosure status, an access control list (ACL) file was designed based on the HL7 clinical document architecture header. The PDF file and its ACL file were transiently stored in the Japanese standard repository, SS-MIX2. The doctor of the designated CDO was then able to access the documents according to the information in the ACL file. Results: From March 2017 to September 2017, 308 documents of 20 patients were disclosed to 3 CDOs using this document-based EHR system. These documents included examination reports, agreement forms concerning the EHR, progress notes, summaries and surgical reports generated from five different systems. Conclusions: The document-based EHR system was able to reveal clinical documents from the EMR to medical staff under a controlled disclosure environment.

Issues related to ensuring patient privacy and data ownership in clinical repositories prevent the growth of translational research. Previous studies have used an aggregator agent to obscure clinical repositories from the data user, and to ensure the privacy of output using statistical disclosure control. However, there remain several issues that must be considered. One such issue is that a data breach may occur when multiple nodes conspire. Another is that the agent may eavesdrop on or leak a user's queries and their results. We have implemented a secure computing method so that the data used by each party can be kept confidential even if all of the other parties conspire to crack the data. We deployed our implementation at three geographically distributed nodes connected to a high-speed layer two network. The performance of our method, with respect to processing times, suggests suitability for practical use.

Because drug‐induced interstitial pneumonia (DIP) is a serious adverse drug reaction, its quantitative risk with individual medications should be taken into due consideration when selecting a medicine. We developed an algorithm to detect DIP using medical record data accumulated in a hospital. Chest computed tomography (CT) is mainly used for the diagnosis of IP, and chest X‐ray reports, KL‐6, and SP‐D values are used to support the diagnosis. The presence of IP in the reports was assessed by a method using natural language‐processing, in which IP was estimated according to the product of the likelihood ratio of characteristic keywords in each report. The sensitivity and the specificity of the method for chest CT reports were 0.92 and 0.97, while those for chest X‐ray reports were 0.83 and 1, respectively. The occurrence of DIP was estimated by the patterns of presence of IP before, during, and after the administration of the target medicine. The occurrence rate of DIP in cases administered Gefitinib; Methotrexate (MTX); Tegafur, Gimeracil, and Oteracil potassium (TS‐1); and Tegafur and Uracil (UTF) was 6.0%, 2.3%, 1.4%, and 0.7%, respectively. The estimated DIP cases were checked by having the medical records independently reviewed by medical doctors. By chart review, the positive predictive values of DIP against Gefitinib, MTX, TS‐1, and UFT were 69.2%, 44.4%, 58.6%, and 77.8%, respectively. Although the cases extracted by this method included some that did not have DIP, this method can estimate the relative risk of DIP between medicines.

Drug-induced interstitial pneumonia (DIP) is a serious adverse drug reaction. The occurrence rete of DIP was evaluated by clinical trial before available in the market. However, due to limited number of cases in clinical trials, it may be inapplicable to the real market. We aimed to seek a method to evaluate the occurrence rate of DIP using clinical data warehouse at a hospital. Initially we developed a method that assesses whether presence of IP was written in reports by natural language processing. Next we detected DIP by estimating IP before, during and after the drug administration. Presence of IP was determined according to the reports of CT if CT was performed, otherwise it was determined based on the changes in the results of chest X-ray, level of KL-6 or SP-D. DIP was determined according to the pattern of presence of IP in each phase. In this study we chose amiodarone as a target drug. The number of patients who suffered from IP caused by amiodarone was 16 (3.9 %), including one definitively diagnosed and 15 strong doubt cases. Most of them could be validated by medical record chart. Using this method, we were able to successfully detect occurrence of DIP from accumulated data in a hospital information system.

Adverse events are detected by monitoring the patient's status, including blood test results. However, it is difficult to identify all adverse events based on recognition by individual doctors. We developed a system that can be used to detect hematotoxicity adverse events according to blood test results recorded in an electronic medical record system. The blood test results were graded based on Common Terminology Criteria for Adverse Events (CTCAE) and changes in the blood test results (Up, Down, Flat) were assessed according to the variation in the grade. The changes in the blood test and injection data were stored in a database. By comparing the date of injection and start and end dates of the change in the blood test results, adverse events related to a designated drug were detected. Using this method, we searched for the occurrence of serious adverse events (CTCAE Grades 3 or 4) concerning WBC, ALT and creatinine related to paclitaxel at Osaka University Hospital. The rate of occurrence of a decreased WBC count, increased ALT level and increased creatinine level was 36.0%, 0.6% and 0.4%, respectively. This method is useful for detecting and estimating the rate of occurrence of hematotoxicity adverse drug events.

Recently one patient received care from several hospitals at around the same time. When the patient visited a new hospital, the new hospital's physician tried to get patient information the previous hospital. Thus, patient information is frequently exchanged between them. Many types of healthcare facilities have implemented an electronic medical record system, but in Japan, healthcare information exchange is often done by paper. In other words, after a clinical doctor prints a referral document and sends it to another hospital's physician, another hospital's doctor receives it and scans to store the EMR in his own hospital's system. It is a wasteful way to exchange healthcare information about a patient. In order to solve this problem, we have developed a cross-institutional document exchange system using clinical document architecture (CDA) with a virtual printing method.

Summary Background: The International Classification of Functioning, Disability and Health (ICF) has been available as a means of coding life functions but the coding process is cumbersome due to the large number of ICF codes. In the current study, we developed ICF code selection tools to support the coding of activity and participation data recorded in domiciliary mental health care reports. Methods: We first developed a search system to facilitate the selection of ICF codes by tracking back through codes’ conceptual trees using a directory tool. We performed a morphological analysis on the training data set to correlate nouns with the ICF codes and obtained an analysis corpus to which numerical scores representing the frequencies of associated ICF codes for each noun were assigned. Based on the obtained corpus we developed a full-text search tool, which could simplify ICF coding relative to that performed using the directory tool. We then evaluated the usefulness of the former tool on the test data set. Results: Using the full-text search tool, correct ICF codes were recorded in the first candidate list for only 54.2% of sentences. However, correct ICF codes appeared on the combined candidate lists for 90.1% of sentences and on the top three candidate lists for 71.7%. In a specific case (General Tasks and Demands), 100% of the correct codes were included on the combined candidate lists. Conclusion: We developed selection tools that effectively supported ICF coding, although it was impossible to fully automate ICF coding. This indicated that ICF codes could more effectively be applied to mental health care.

This system is designed as a tool which solves some challenges imposed in community mental health care. Using a computer network, the electronic-bulletin-board-like system which uses ICF-based data input is applied, for the purpose of sharing community mental health care consumer information. Such system is expected to facilitate health care consumer rehabilitation and in the prevention of a recurrence of an illness. It will become a vital necessity in the future of community mental health care.

This paper proposes an interface for wearable computers using augmented reality (AR) environment which allows the user to handle virtual objects with his/her own hands intuitively. The proposed system is constructed with a wearable computer and a head mounted display (HMD). A pair of cameras is attached to HMD positioned as a pair of eyes. Using the stereo measurement, the finger's postures are captured and displayed together with virtual objects on a video-see-though HMD. The system tracks the fingertips motion of the index finger and the thumb so as the user can pick up, move and rotate the virtual object as if treating a real object with the hands in the real world. While the stereo measurement, range information is added to the region of the hand Visual occlusion is synthesized between the user's hand and the virtual objects including 2D based graphical user interfaces, 3D computer graphics models and so on.

The cyclic micro-deformation of the arterial wall with pulsatile flow was measured to get fundamental data for estimation of the mechanical stress in endothelial cells. The descending aorta (1-2 mm diameter) of an anesthetized rat was exposed under thoracotomy. The displacement measuring system was assembled with the charge coupled devise (CCD) of laser sensor. The movement of laser beam (670 nm wave length) reflected at the vessel wall was calibrated to the movement of the arterial wall. The fluctuating movement was also measured at four points marked on the vessel wall with CCD camera to distinguish the circumferential movement from the longitudinal one. The results showed that the present designed system has enough resolution to measure the arterial vessel wall cyclic-micro-fluctuation, which is 10 percent of diameter in the circumferential direction without deformation in the longitudinal direction with the cardiac beating in vivo.