ABSTRACT
The purpose of this study was to evaluate the accuracy of Japanese fill-in-the-blank questions in the radiological technology field using six BERT models. Our original model, created using 330,000 abstracts related to radiological technology, was as accurate as other models created by Wikipedia when vocabulary size was the same. Further improvements in accuracy can be expected if we can increase the vocabulary size and add Japanese Wikipedia data in training.
Subject(s)
Technology, Radiologic , JapanABSTRACT
The purpose of this study was to demonstrate Japanese radiographic examination codes JJ1017 in establishing typical values for a wide variety of general radiography. About 200,000 sets of examination data were collected, including exposure conditions, JJ1017 code applied, examination room numbers and patient information. Typical values for adults, children, and infants were calculated from the collected data, and the following items were examined: comparing typical values of general radiography in Japan DRLs 2015 and typical values in a facility; comparison of typical values between X-ray equipment for examinations of DRLs 2015; comparison of typical values for different procedures at the same anatomical site; identification of examination items associated with high radiation doses. The total numbers of JJ1017 codes applicable to the examinations were 45,372 for adults, 542 for children, and 2339 for infants. To calculate the typical values and compare these with the DRLs, we used a combination of JJ1017 anatomical codes, posture codes, and direction of radiation codes. The combination of these codes allowed the calculation of a typical value and comparison with DRLs 2015. Comparison between devices reveals differences in radiation doses and provides an opportunity to review the characteristics of the devices and their operation to suggest dose reductions. By calculating typical values for examination items for which the DRLs were not available, we were able to identify examination items with high doses in a facility and suggest items that should be audited in the facility.
Subject(s)
Radiation Dosage , Child , Adult , Humans , Japan , Radiography , X-Rays , Reference ValuesABSTRACT
The purpose of this study was to evaluate the accuracy of deep neural machine translation focused on medical device adverse event terminology. 10 models were obtained, and their English-to-Japanese translation accuracy was evaluated using quantitative and qualitative measures. No significant difference was found in the quantitative index except for a few pairs. In the qualitative evaluation, there was a significant difference and googletrans and GPT-3 were regarded as useful models.
Subject(s)
Artificial Intelligence , Equipment Failure , Translating , Terminology as TopicABSTRACT
The terminology in radiological technology is crucial, encompassing a broad range of principles from radiation to medical imaging, and involving various specialists. This study aimed to evaluate the accuracy of automatic synonym detection considering the characteristics of the Japanese language by Word2vec and fastText in the radiological technology field for the terminology elaboration. We collected around 340 thousand abstracts in Japanese. First, preprocessing of the abstract data was performed. Then, training models were created with Word2vec and fastText with different architectures: continuous bag-of-words (CBOW) and skip-gram, and vector sizes. Baseline synonym sets were curated by two experts, utilizing terminology resources specific to radiological technology. A term in the dataset input into the generated models, and the top-10 synonym candidates which had high cosine similarities were obtained. Subsequently, precision, recall, F1-score, and accuracy for each model were calculated. The fastText model with CBOW at 300 dimensions was most precise in synonym detection, excelling in cases with shared n-grams. Conversely, fastText with skip-gram and Word2vec were favored for synonyms without common n-grams. In radiological technology, where n-grams are prevalent, fastText with CBOW proved advantageous, while in informatics, characterized by abbreviations and transliterations, Word2vec with CBOW was more effective.
ABSTRACT
PURPOSE: The purpose of this study was to assess inter-rater reliability and workload for creating accurate training data in the clinical evaluation of mammographic positioning for deep learning. METHODS: A total of 107 mammographic images without lesions were labeled by two certified radiologic technologists in seven items: six clinical image evaluation criteria in positioning and breast tissue density. The kappa coefficient was calculated as an indicator of interrater reliability. Furthermore, the labeling cost per image was calculated based on labeling time and salary for the technologists. RESULTS: The kappa coefficients were 0.71 for inframammary fold, 0.43 for nipple in profile, 0.45 for great pectoral muscle, 0.10 for symmetrical images, and 0.61 for retromammary fat. No significant difference was found in the coefficients of spread of breast tissue. The cost per image was calculated at 11.0 yen. CONCLUSION: The inter-rater reliability for the inframammary fold, nipple in profile, great pectoral muscle, and retromammary fat ranged from "moderate" to "substantial." The reliability for symmetrical images was "slight," indicating the need for a consensus among evaluators during labeling. The labeling cost was equivalent to or higher than that of existing services.
Subject(s)
Adipose Tissue , Mammography , Reproducibility of Results , Certification , Costs and Cost AnalysisABSTRACT
The purpose of our study is to identify the patterns of vectors in similar/dissimilar pairs of definition sentence created by Word2vec and doec2vec for elaboration of the terminology for Japanese Medical Device Adverse Events. 2-dimension vector space created by t-SNE showed that the pair with true positive located closer in a vector space, especially Doc2vec had a strong tendency. Comparing with Word2vec, Similar vectors in Doc2vec were close and tended to form clusters.
Subject(s)
Language , JapanABSTRACT
The image management function of picture archiving and communication system (PACS) plays a very important role while electronically preserving medical records. However, details of these mechanisms vary from manufacturer to manufacturer, which could be a problem if system specifications or image management functions are not confirmed by a user during system installation or operation. The purpose of this research was to record the differences in PACS settings and image data management in hospitals. A questionnaire survey was conducted in 261 hospitals having the PACS system in Hokkaido prefecture, Japan. Questionnaire contents consisted of a face sheet and the following two items: 1) PACS image deletion method, 2) rules regarding the PACS receiving an image of data that was already registered. Results of the questionnaire survey showed that the image deletion method in the PACS server, and the overwriting rule in case of existent images being re-transmitted to the server were different between different hospitals. Furthermore, these variations were not dependent on the vendors or manufacturers, but on the settings of each hospital. Since the image management method of PACS is not standardized, we were considered to be needed more attention and appropriate regulation for safe management of electronic medical records as per the national guideline.
Subject(s)
Radiology Information Systems , Computers , Information Management , Japan , Surveys and QuestionnairesABSTRACT
BACKGROUND: For standardization of terms in the reports of medical device adverse events, 89 Japanese medical device adverse event terminologies were published in March 2015. The 89 terminologies were developed independently by 13 industry associations, suggesting that there may be inconsistencies among the terms proposed. The purpose of this study was to integrate the 89 sets of terminologies and evaluate inconsistencies among them using SPARQL. METHODS: In order to evaluate the inconsistencies among the integrated terminology, the following six items were evaluated: (1) whether the two-layer structure between category term and preferred term is consistent, (2) whether synonyms of a preferred term are involved. Reversing the layer-category order of matching was also performed, (3) whether each preferred term is subordinate to only one category term, (4) whether the definitions of terms are uniquely determined, (5) whether CDRH-NCIt terms corresponding to preferred terms are uniquely determined, (6) whether a term in a medical device problem is used for patient problems. RESULTS: About 60% of the total number of duplicated terms were found. This is because industry associations that created multiple terminologies adopted the same terms in terminologies of similar medical device groups. In the case that all terms with the same spelling have the same concept, efficient integration can be achieved automatically using RDF. Furthermore, we evaluated six matters of inconsistency in this study, terms that need to be reviewed accounted for about 10% or less than 10% in each item. CONCLUSIONS: The RDF and SPARQL were useful tools to explore inconsistencies of hierarchies, definition statements, and synonyms when integrating terminolgy by term notation, and these had the advantage of reducing the physical and time burden.
Subject(s)
Language , Humans , JapanABSTRACT
PURPOSE: This study was designed to assess working environment preferences of students in the Department of Radiological Technology using conjoint analysis for establishing an efficient medical system. METHOD: We carried a questionnaire survey on working environment preferences for 196 students in the Department of Radiological Technology in Japan. We defined eight characteristics for virtual medical facilities as follows: presence of colleagues who can be consulted, employment status, number of night shift per month, academic meeting participation, number of hospital beds, possession of nuclear medicine imaging systems and radiation therapy systems, location of medical facilities, and change rate in annual income. A total of 18 virtual medical facilities were selected by an orthogonal array table using above-mentioned characteristics. The acquired data by the pairwise comparison method were analyzed by conjoint analysis. Marginal rates of substitution that represent students' preferences were also calculated. RESULT: The factors that influenced their preferences were the following: placement of medical facilities in great city, presence of colleagues who can be consulted, employment status is not non-regular employment, set up of nuclear medicine imaging systems and radiation therapy systems, the number of night shift is twice per month, and attendances at academic meetings. CONCLUSION: In summary, students in the Department of Radiological Technology tend to prefer the facilities with regular employment, great city, presence of colleagues who can be consulted, and possession of nuclear medicine imaging systems and/or radiation therapy systems.
Subject(s)
Technology, Radiologic , Workplace , Humans , Japan , Students , Surveys and QuestionnairesABSTRACT
PURPOSES: The purposes of this study were to automatically extract full forms from abbreviations by using Word2vec for terminology expansion and determine the optimal parameters that ensure the highest accuracy. METHODS: Approximately 300000 English abstracts on "image diagnosis" were collected using PubMed from January 1994 to December 2018. As preprocessing, all uppercase letters in the collected data were converted to lowercase letters, and symbols were deleted. In addition, compound word recognition was performed using RadLex published by the Radiological Society of North America and the abbreviation collection published by the Japanese Society of Radiological Technology. Next, distributed representations were generated by two algorithms, continuous bag-of-words (CBOW) and Skip-gram, by using the following parameters: iteration numbers (3-85) and dimensions of word vectors (50-1000). Abbreviations were input to the generated distributed representations, and full forms with the highest cosine similarities with the abbreviations were identified. Then, the rates of the correct answers were calculated by comparing the predicted full forms to 214 gold standards extracted from the abbreviation collection. RESULTS: The highest correct answer rate was 74.3% by Skip-gram, 200 dimensions and 10 iterations. This rate was higher in Skip-gram than in CBOW for all the tested conditions. CONCLUSION: The accuracy of extracting the full forms by Word2vec is 74.3%, and this result contributes to the consistency of a terminology and the efficiency of terminology expansion.
Subject(s)
Algorithms , Health ServicesABSTRACT
BACKGROUND: Public interest in radiation rose after the Tokyo Electric Power Company (TEPCO) Fukushima Daiichi Nuclear Power Station accident was caused by an earthquake off the Pacific coast of Tohoku on March 11, 2011. Various reports on the accident and radiation were spread by the mass media, and people displayed their emotional reactions, which were thought to be related to information about the Fukushima accident, on Twitter, Facebook, and other social networking sites. Fears about radiation were spread as well, leading to harmful rumors about Fukushima and the refusal to test children for radiation. It is believed that identifying the process by which people emotionally responded to this information, and hence became gripped by an increased aversion to Fukushima, might be useful in risk communication when similar disasters and accidents occur in the future. There are few studies surveying how people feel about radiation in Fukushima and other regions in an unbiased form. OBJECTIVE: The purpose of this study is to identify how the feelings of local residents toward radiation changed according to Twitter. METHODS: We used approximately 19 million tweets in Japanese containing the words "radiation" (), "radioactivity" (), and "radioactive substances" () that were posted to Twitter over a 1-year period following the Fukushima nuclear accident. We used regional identifiers contained in tweets (ie, nouns, proper nouns, place names, postal codes, and telephone numbers) to categorize them according to their prefecture, and then analyzed the feelings toward those prefectures from the semantic orientation of the words contained in individual tweets (ie, positive impressions or negative impressions). RESULTS: Tweets about radiation increased soon after the earthquake and then decreased, and feelings about radiation trended positively. We determined that, on average, tweets associating Fukushima Prefecture with radiation show more positive feelings than those about other prefectures, but have trended negatively over time. We also found that as other tweets have trended positively, only bots and retweets about Fukushima Prefecture have trended negatively. CONCLUSIONS: The number of tweets about radiation has decreased overall, and feelings about radiation have trended positively. However, the fact that tweets about Fukushima Prefecture trended negatively, despite decreasing in percentage, suggests that negative feelings toward Fukushima Prefecture have become more extreme. We found that while the bots and retweets that were not about Fukushima Prefecture gradually trended toward positive feelings, the bots and retweets about Fukushima Prefecture trended toward negative feelings.
Subject(s)
Emotions/physiology , Fukushima Nuclear Accident , Social Media/standards , Attitude , Female , Humans , Japan , Male , Surveys and QuestionnairesABSTRACT
The purpose of this study was to investigate the trends of researches regarding radiological technology. We collected research papers published from 2007 to 2017 from Japanese Society of Radiological Technology (JSRT). After preprocessing, we performed morphological analysis using terminology from Japan Radiological Society, Japan Society of Medical Physics, and JSRT to extract technical terms. Furthermore, we calculated the Jaccard similarity coefficient to represent the similarity between two terms. This value ranged from 0 to 1, where 0 implied that the terms were completely dissimilar. Finally, in order to detect terms that characteristically appear in each year, we visualized co-occurring terms by using network diagrams. From the morphological analysis, 5471 technical terms were extracted. The most frequency term was "image" from 2007 to 2017. "Phantom" and "CT" were frequent terms after "image." In addition, the number of research papers including "image," "phantom," and "CT" were increasing. For network analysis, the characteristic terms in 2007 were "filter" and "HU"; those in 2012 were "dimension," "standard deviation,"and "artifact"; and those in 2017 were "PET," "scattered ray," and "collimator." In conclusion, the highest interest research topic in radiological technology was "image," and recently, there has been a tendency to be interested in topics related to nuclear medicine.
Subject(s)
Data Mining , Technology, Radiologic , Japan , Phantoms, Imaging , Radionuclide ImagingABSTRACT
A common cause for image retakes in mammography is lack of proper positioning. Image retake is a potential hazard of increased radiation exposure to patient, patient discomfort and pain. Therefore, a mammographer has to provide fast exams with fewer retakes. Although evaluation of how a mammographer is effectively positioning has been studied in many ways, little research has been conducted to analyze visual attention. In this study, eye tracking system was adopted to detect eye movements and locations within a participants' visual fields during positioning. Eye tracking system has been widely used for assessing technical skills and risk awareness, and for comparing the skills between experts and novices. Two skilled mammographers and two novices were recruited. Positioning related activities were divided into two phases: patient's observation and breast positioning. Breast positioning was sub-divided into three stages to compare visual attention between experts and novices using heatmap and gaze plot. Although the expert tended to check each point, the novices tended to have a relatively short gaze of the outer breast region was observed. In the future, a comparative evaluation using clinical images is necessary; however, the eye tracking system to visualize attention contributes to medical safety during positioning.
Subject(s)
Eye Movements , Mammography , Attention , Breast , HumansABSTRACT
The purpose of this study is to extract similar term definitions used in the terminology of Japanese medical device adverse events. We employed Levenshtein and Jaro-Winkler distances as edit distances and Skip-gram, continuous-bag of words, and fast text to produce distributed representations in Word2Vec. A comparison of the accuracies of the models showed that Levenshtein distance had higher specificity whereas Skip-gram had higher sensitivity as compared to the other models.
Subject(s)
Equipment and Supplies/adverse effects , Terminology as TopicABSTRACT
We investigated problems concerning medical imaging management using PACS in medical settings through text analysis. We conducted a questionnaire survey in Hokkaido, Japan, where PACS related problems were described by radiological technologists. After analyzing the descriptions in response to the questionnaire, we detected four main problems in PACS using a word co-occurrence network; i) image management, ii) image fetching error, iii) DICOM violation and iv) name notation.
Subject(s)
Natural Language Processing , Radiology Information Systems , Information Management , JapanABSTRACT
The purpose of this research was to record the differences in picture archiving and communication system (PACS) settings in hospitals. A questionnaire survey was conducted among 261 hospitals that used the PACS system in Japan. As a result, the image deletion method in the PACS server, and other rules were different among hospitals. Furthermore, these variations were dependent on settings of each medical institution. It seemed necessary to define standardized rules for safe image management.
Subject(s)
Radiology Information Systems , Computers , Information Management , Japan , Surveys and QuestionnairesABSTRACT
BACKGROUND: The Great East Japan Earthquake on March 11, 2011, triggered a huge tsunami, causing the Fukushima Daiichi nuclear disaster. Radioactive substances were carried in all directions, along with the risks of radioactive contamination. Mass media companies, such as television stations and news websites, extensively reported on radiological information related to the disaster. Upon digesting the available radiological information, many citizens turned to social media, such as Twitter and Facebook, to express their opinions and feelings. Thus, the Fukushima Daiichi nuclear disaster also changed the social media landscape in Japan. However, few studies have explored how the people in Japan who received information on radiation propagated the information. OBJECTIVE: This study aimed to reveal how the number of tweets by citizens containing radiological information changed regionally on Twitter. METHODS: The research used about 19 million tweets that included the terms "radiation," "radioactivity," and "radioactive substance" posted for 1 year after the Fukushima Daiichi nuclear disaster. Nearly 45,000 tweets were extracted based on their inclusion of geographic information (latitude and longitude). The number of monthly tweets in 4 districts (Fukushima Prefecture, prefectures around Fukushima Prefecture, within the Tokyo Electric Power Company area, and others) were analyzed. RESULTS: The number of tweets containing the keywords per 100,000 people at the time of the casualty outbreak was 7.05 per month in Fukushima Prefecture, 2.07 per month in prefectures around Fukushima Prefecture, 5.23 per month in the area within Tokyo Electric Power Company, and 1.35 per month in others. The number of tweets per 100,000 people more than doubled in Fukushima Prefecture 2 months after the Fukushima Daiichi nuclear disaster, whereas the number decreased to around 0.7~0.8 tweets in other districts. CONCLUSIONS: The number of tweets per 100,000 people became half of that on March 2011 3 or 4 months after the Fukushima Daiichi Nuclear Plant disaster in 3 districts except district 1 (Fukushima Prefecture); the number became a half in Fukushima Prefecture half a year later.