BACKGROUND:We aimed to construct an artificial intelligence-based model for detecting oral cancer and dysplastic leukoplakia using oral cavity images captured with a single-lens reflex camera. SUBJECTS AND METHODS:We used 1043 images of lesions from 424 patients with oral squamous cell carcinoma (OSCC), leukoplakia, and other oral mucosal diseases. An object detection model was constructed using a Single Shot Multibox Detector to detect oral diseases and their locations using images. The model was trained using 523 images of oral cancer, and its performance was evaluated using images of oral cancer (n = 66), leukoplakia (n = 49), and other oral diseases (n = 405). RESULTS:For the detection of only OSCC versus OSCC and leukoplakia, the model demonstrated a sensitivity of 93.9% versus 83.7%, a negative predictive value of 98.8% versus 94.5%, and a specificity of 81.2% versus 81.2%. CONCLUSIONS:Our proposed model is a potential diagnostic tool for oral diseases.

Adverse events after surgery not only affect the patient’s recovery but also increase the burden on doctors and patients due to prolonged hospitalization. Predicting adverse events from patient data before surgery with a machine learning method is highly expected. It is difficult to collect a large amount of patient data since the number of surgeries in a year is limited and predict the occurrence of adverse events accurately since patient data are imbalanced data. To improve the accuracy of adverse event prediction, this paper proposes data generation with Filtered $\beta $ -VAE for the preoperative prediction of adverse events. Filtered $\beta $ -VAE has filters by the reconstruction error and by a machine learning method. After $\beta $ -VAE generates minority class data, the two layers of filtering are used to remove low-quality minority class data that have little contribution to the adverse event prediction. In the evaluations, patient data obtained from Tokyo Dental University Ichikawa General Hospital were used. The proposed method can predict adverse events with a recall of 0.848, which is 5.6% more accurate than existing methods. The effects of filtering in Filtered $\beta $ -VAE are visualized, and the reasons for the improvement in prediction accuracy are clarified. Furthermore, this paper shows that $\beta $ -VAE can generate arbitrary patient data even in table data, corresponding to the distribution of the original patient data.

In the last few years, artificial intelligence (AI) research has been rapidly developing and emerging in the field of dental and maxillofacial radiology. Dental radiography, which is commonly used in daily practices, provides an incredibly rich resource for AI development and attracted many researchers to develop its application for various purposes. This study reviewed the applicability of AI for dental radiography from the current studies. Online searches on PubMed and IEEE Xplore databases, up to December 2020, and subsequent manual searches were performed. Then, we categorized the application of AI according to similarity of the following purposes: diagnosis of dental caries, periapical pathologies, and periodontal bone loss; cyst and tumor classification; cephalometric analysis; screening of osteoporosis; tooth recognition and forensic odontology; dental implant system recognition; and image quality enhancement. Current development of AI methodology in each aforementioned application were subsequently discussed. Although most of the reviewed studies demonstrated a great potential of AI application for dental radiography, further development is still needed before implementation in clinical routine due to several challenges and limitations, such as lack of datasets size justification and unstandardized reporting format. Considering the current limitations and challenges, future AI research in dental radiography should follow standardized reporting formats in order to align the research designs and enhance the impact of AI development globally.

Intraoperative hypotension may occur during surgery, depending on the patient's condition and the amount of anesthetics administered. Intraoperative hypotension causes adverse events such as myocardial infarction, acute kidney injury, and stroke, which may affect patient prognosis and should be avoided. However, it is challenging to predict intraoperative hypotension before surgery. Therefore, the authors propose a system that predicts the occurrence of intraoperative hypotension with higher accuracy than existing studies to help preventing intraoperative hypotension. Specifically, the occurrence of intraoperative hypotension is predicted by using information about personal lifestyle and medical history obtained from the preanesthesia questionnaire in addition to the patients' baseline characteristics such as age and gender that have already been used. The authors constructed prediction model, that predicted intraoperative hypotension with an F-measure of 0.724.

Electronic anesthesia record data have been accumulated, and efforts to solve medical problems using data analysis methods and machine learning have been conducted. Post-induction hypotension frequently occurred after induction of anesthesia. Intraoperative hypotension is associated with various adverse events such as myocardial infarction and cerebral infarction. In a related study, eight machine learning methods were used to construct hypotension prediction models and evaluated by area under the curve (AUC), using data collected from an institution in the United States. Nevertheless, it was not focused on improving prediction power. This paper aims to predict post-induction hypotension with high prediction power using 1,626 electronic anesthesia record data. Our hypotension prediction model using a stacking method is introduced. F-measure 0.60 was achieved by using our method through the evaluation.

Post-induction hypotension frequently occurred after anesthesia induction. Avoiding post-induction hypotension is important as it is associated with postoperative adverse outcomes. Related studies have shown that the dose of anesthetic induction drugs affects the post-induction hypotension. The purpose of this study is to propose an anesthetic dose that does not cause post-induction hypotension according to the patient's condition. A model for predicting the optimal dose of an anesthetic induction drug is constructed using a regression model which is one of machine learning methods by focusing on electronic anesthesia records. The prediction coefficient of determination 0.5008 was achieved by adjusting the explanatory variables and parameters and using ridge regression.

The authors aim to develop a system that saves money by providing tips (proposed actions; the tips are called know-how) based on user preferences. In order to select helpful tips, this study develops a method that improves estimation accuracy by optimizing the tip indices that represent the tendency in user preferences. Concretely, the method optimizes the tip indices using both the result generally acquired by five-level evaluations and the result gradually acquired by a paired comparison method in which the users are requested to select one of pair of tips. Evaluation experiments show the general trend that estimation accuracy rises with the amounts of paired comparison data.

This paper proposes a method that predicts customer value by focusing on purchasing behavior. The method generates a relevance model for purchase days and amount in each period between customer value and purchasing histories beforehand based on a consumer panel survey. The authors adopt the random forest method to generate the prediction model. The proposed method facilitates the provisioning of smart customer management to each customer according to level such as suggesting products or services.

This paper develops a prototype implementation of our proposed unconscious participatory sensing system and evaluates its performance about a measurement process on a real smartphone device. The proposed system consists of various beacon devices and smartphone devices. It also requests smartphone owners to install a special application on their smartphones to collect measurement information from beacon devices. Each beacon device has Bluetooth low energy (BLE) module to communicate with a smartphone device. Hence, the proposed system can collect measurement information from beacon devices through smartphones. Additionally, beacon devices can work by a battery for a long time because BLE is appropriate for a low power operation. The feature of the proposed scheme is to activate a special smartphone application in a sleep state to upload measurement information. Therefore, it does not require smartphone owners to launch the smartphone application to collect measurement information. We employ the iBeacon function which is the special beacon mechanism for iOS to active the application. In the experimental evaluation, we have developed a beacon device with a Raspberry Pi and a special application for iOS. Since a background operation period of iOS applications is limited due to the power saving mechanism on iOS, we have evaluated the measurement process of the proposed system. From the results, we can find that the developed application for iOS can collect information from beacon devices in a background operation even if the background operation period is limited.

This paper proposes a method to estimate consumer’s lifestyle by focusing on micro purchasing behavior. The method acquires relevance model for each product category between lifestyles and purchasing histories with detailed product ID (JAN code) beforehand based on consumer panel survey. The model acquisition employs Neural Network, a machine learning method. Appropriate relevance models are adaptively selected for each lifestyle and combined to achieve optimal performance on lifestyle prediction. The method is useful to realize smart recommendation such as suggesting products with reflection of each consumer’s lifestyle.

This paper proposes a method that estimatesconsumer family-structure attributesby focusing on purchasing behavior. The method generatesa relevance model for each product type between family-structure attributes and purchasing histories beforehand based on a consumer panel survey. Random Forest, a machine learning method, is employed to generate the model. The proposed method facilitates provisioning of smart recommendations to the consumer family such as suggesting products that reflectthe family structure attributes of the consumer.

This paper proposes a method to estimate the preference of customers based on store check-in histories. The proposed method can distinguish the preferences of customers who have no purchase histories. We adopt a machine learning algorithm for model acquisition. The estimation results can improve the efficiency of visitor promotion campaigns and advertising campaigns. An actual visitor promotion trial indicates the effectiveness of the proposed method.

This paper proposes a beacon device based unconscious participatory sensing system consisting of beacon devices, smartphone devices, and management servers. Beacon devices installed at measurement locations find a neighbor smartphone device to request to relay measurement information to a management server or to request measurement with builtin sensors. The proposed processes can be carried out without participators’ interaction behavior. Hence, the proposed system just requires participators to install the special smartphone application, and participators can easily join a sensing network and can work for participatory sensing unconsciously. We employ the Bluetooth Low Energy (BLE), which is a short-range wireless communication standard, to realize communication between the beacon device and the smartphone application. Additionally, we achieve low power operations of the smartphone application by employing iBeacon, which is a shortrange recognition mechanism for BLE. To the best of our knowledge, this is the first attempt to propose the unconscious participatory sensing system with smartphone devices and BLE based beacon devices. From the experimental implementation with a Raspberry Pi and an Android OS 4.4 based smartphone, we have confirmed that the proposed signaling works well with the consumer devices.

The objective of this research is to realize a method that detects a user's entrance to predefined geographic area so that appropriate services are provided without explicit user action. Though such geofencing technology is gathering attention due to its wide range of applications, it is a challenge to overcome trade-off between power consumption and detection accuracy. Conventional variable interval positioning methods fall short on battery saving due to difficulties such as sparse and erroneous position measurement data, and the unpredictable speed and trajectory of the terminal. In this paper, we propose a method for position detection whose activation frequency is determined by speed toward the target spot. The method is robust against positioning error and fluctuation of the terminal's movement by leveraging the access angle to the target spot. Simulation results show that the proposed access speed correction method reduces power consumption from 42% to 52% against conventional method, while holding the false negative ratio to detect the target spot to less than 5%.

Operation history middleware collects various operation data on smartphones. Because of privacy concerns about sensitive operation data, it needs to meet fine-grained collection needs including collection based on temporal-spatial conditions and temporary suppression of the collection of specific operation types. The on-demand collection method proposed herein dynamically determines the minimum collection set of operation types and their storage period. We develop an Android-based prototype that implements five collection techniques and a selective encryption function to protect sensitive data in the local database. Our evaluation shows that the collection techniques incur low overhead. Encryption overhead in recording is less than 26 milliseconds.

Abstract The availability, distribution, and metabolism of four typical selenium sources [inorganic selenite and selenate, and organic selenomethionine (SeMet) and methylselenocysteine (MeSeCys)] were compared by administering them simultaneously through a parenteral route. The four selenium sources were each labeled with a different enriched selenium isotope (82Se, 78Se, 77Se, and 76Se, respectively), and administered intravenously at the dose of 25 μg Se/kg body weight each to rats that had been depleted of natural abundance selenium with a single isotope, 80Se, by feeding 80Se-selenite in drinking water and a selenium-deficient diet. At 1 h post-injection, the amounts of the four tracers recovered from major organs and blood comprised around 70, 55, and 50 % of the doses for selenite, MeSeCys and SeMet, and selenate, respectively, being most abundant in the liver. The intact precursors, except for selenite, were recovered from all organs. 77Se and 76Se of SeMet and MeSeCys origin, respectively, were much more efficiently recovered from the pancreas than selenite and selenate, in forms mostly bound to proteins together with intact forms, suggesting that SeMet and MeSeCys are preferentially distributed directly to the pancreas. The incorporations of selenium into selenoprotein P (Sel P) and selenosugars were most efficient from selenite and less efficient from SeMet, suggesting that selenite was most efficiently utilized for the syntheses of selenoproteins and selenosugars. Although selenate was partly excreted into the urine in its intact form, it was retained longer in the plasma in its intact form than the other selenium sources. The advantage of simultaneous administration of multiple precursors each labeled with a different enriched isotope to depleted hosts followed by simultaneous tracing of the labeled isotopes over the conventional method with a single tracer is emphasized together with cautions that may occur with the new multiple tracer method.

We performed a retrospective study to examine the protective effect of low-dose dexamethasone (DEX) on delayed adverse events induced by carboplatin (CBDCA)-based combination chemotherapy in patients with thoracic tumors. Low-dose DEX (4-8 mg/day) was administered on day 1 and after, in addition to a serotonin 5-HT3 receptor antagonist. The acute adverse events (day 1) were well controlled in the patients with or without co-treatment of DEX. On the other hand, the delayed nausea, emesis, anorexia, and fatigue after day 2 failed to be controlled by 5-HT3 antagonist alone. Co-treatment with DEX significantly suppressed the grade of the delayed adverse events during days 2-10. The mean ratio of complete protection during days 2-10 were significantly higher in the DEX-treated group compared with the non-DEX-treated group. These results reveal that low-dose DEX is a clinically effective treatment for the prevention of delayed adverse events induced by CBDCA-based combination chemotherapy.

Monitoring the adverse reaction patterns specific to individual patients is important to avoid subsequent reactions. Gynecologic cancer chemotherapy is often implemented repeatedly with an altered protocol during prolonged terms. The purpose of this study was to develop and assess the efficacy of a worksheet that pharmacists can use to analyze adverse reaction patterns in individual patients with gynecologic chemotherapy. The worksheet which we developed consisted of multiple sections. One section is for necessary drug information for the proper use of antineoplastic agents. Another section is for the following items recorded by the pharmacists: a) patients' basic information such as stage of disease and protocol, b) state of implementation and break of chemotherapy and supportive therapy on calendar, and c) laboratory data and symptoms. We arranged the last item below the calendar and enabled pharmacists to easily assess individual adverse reactions coupled with the treatment course. Reviews of the developed worksheet indicated that the worksheet led to the convenient detection of individual adverse reaction patterns and effective prevention of additional adverse reactions. This monitoring sheet covering long-term chemotherapy which was designed to predict individual adverse reaction patterns will improve the individualization and safety of gynecologic chemotherapy.

The purpose of this study was to assess patient participation in cancer therapy and the sharing of patient information among the medical care team (physicians, nurses, pharmacists, and especially patients). We monitored the side effects of cancer chemotherapy with patients, and developed two support tools: One scored the points of subjective symptoms (fatigue, anorexia, nausea, etc) by patients, and the other recorded objective symptoms (clinical examination data) by pharmacists. It is most important that they attend each patient at their bedside. At this time, the trial was evaluated by questionnaire survey by inpatients receiving cancer chemotherapy (n=15). As a result, all patients (15/15) responded that this trial was necessary. This trial addressed the following: 1) increased communication between patients and medical staff concerning side effects (14/15), 2) increased interest in side effects (10/15), 3) when a patient tells medical staff about side effects, they act on it (10/15). None of the patients felt inconvenienced by scoring every day (0/15), or anxiety about side effects (0/15). Furthermore, all patients (15/15) responded that "participation of pharmacists in cancer chemotherapy" was necessary. This trial revealed no problems and suggested that patients related to the center of medical care. We should be careful in interpreting results of this small sized trial; however, the following conclusions should be reached: 1) introduction of monitoring side effects of cancer chemotherapy with patients, 2) develop communication among the medical care team.