Deep learning is a subset of machine learning that can be employed to accurately predict biological transitions. Eliminating hepatitis B surface antigens (HBsAgs) is the final therapeutic endpoint for chronic hepatitis B. Reliable predictors of the disappearance or reduction in HBsAg levels have not been established. Accurate predictions are vital to successful treatment, and corresponding efforts are ongoing worldwide. Therefore, this study aimed to identify an optimal deep learning model to predict the changes in HBsAg levels in daily clinical practice for inactive carrier patients. We identified patients whose HBsAg levels were evaluated over 10 years. The results of routine liver biochemical function tests, including serum HBsAg levels for 1, 2, 5, and 10 years, and biometric information were obtained. Data of 90 patients were included for adaptive training. The predictive models were built based on algorithms set up by SONY Neural Network Console, and their accuracy was compared using statistical analysis. Multiple regression analysis revealed a mean absolute percentage error of 58%, and deep learning revealed a mean absolute percentage error of 15%; thus, deep learning is an accurate predictive discriminant tool. This study demonstrated the potential of deep learning algorithms to predict clinical outcomes.

(1) Background: The impacts of metal artifacts (MAs) on the contouring workload for head and neck radiotherapy have not yet been clarified. Therefore, this study evaluated the relationship between the contouring time of the MAs area and MAs on head and neck radiotherapy treatment planning. (2) Methods: We used treatment planning computed tomography (CT) images for head and neck radiotherapy. MAs were classified into three severities by the percentage of CT images containing MAs: mild (75%). We randomly selected nine patients to evaluate the relationship between MAs and the contouring time of the MAs area. (3) Results: The contouring time of MAs showed moderate positive correlations with the MAs volume and the number of CT images containing MAs. Interobserver reliability of the extracted MAs volume and contouring time were excellent and poor, respectively. (4) Conclusions: Our study suggests that the contouring time of MAs areas is related to individual commitment rather than clinical experience. Therefore, the development of software combining metal artifact reduction methods with automatic contouring methods is necessary to reducing interobserver variability and contouring workload.

Objectives: This study is aimed to perform imaging of phantoms to simulate intraoral ultrasound and to investigate the formation mechanism of mucosal epithelial echo.Methods: In this study, we used phantoms, simulated mucosal layer (SML) with thickness of 0.5, 1.0, 3.0 mm and no SML, and an intraoperative small linear probe as ultrasound transducer. Acoustic coupling polymer gel and agar gel were used as the coupling material, and polyvinylidene chloride (PVDC), polyethylene (PE) and polymethylpentene (PMP) plastic wrap were used. The thicknesses of a surface echo and mucosal epithelial echo were measured on B‐mode images.Results: A surface echo approximately 0.5 mm thick and a hyperechoic line structure approximately 1.0 mm deep in the 3.0‐mm SML, which were presumed to be artifacts, were visualized with all combinations of acoustic coupling material and plastic wrap except with acoustic coupling agar gel and PMP wrap. When using acoustic coupling polymer gel, a hypoechoic line approximately 0.2 mm thick was observed, even when no SML was present. Thickness of the SML appeared to be lesser on ultrasound images than its actual depth, and the most accurate measurement of thickness was obtained using acoustic coupling agar gel together with PMP wrap.Conclusion: When using acoustic coupling polymer gel, artifacts may have prevented accurate visualization over an area that is approximately 0.6–0.7 mm thick. This shows that when intraoral ultrasound is performed by this method, it may not be possible to accurately visualize the mucosal epithelium.

Background/Objectives: Oral squamous cell carcinoma (OSCC) carries a risk of late metastasis not only in advanced stages but also in early stages. In this study, we built and tested radiomics-based machine learning (ML) models for predicting the risk of metastasis from early OSCC on ¹⁸F-FDG positron emission tomography (PET). Methods: Patients diagnosed with T1 or T2 squamous cell carcinoma who underwent a preoperative ¹⁸F-FDG PET-CT examination at a single institution between 2016 and December 2022 were included in this retrospective study. The presence or absence of late cervical lymph node metastasis was confirmed for all patients. Among the radiomics features extracted from the images, we selected those that were useful for predicting late metastasis and used them to create ML models. We then verified the prediction accuracy of the models. Results: A total of 109 subjects were included, of which 31 had late lymph node metastasis and 78 were without metastasis. The most accurate ML model created using radiomics features selected from the subject cases had an area under the curve of 0.977 and accuracy of 87.5%. Conclusions: We confirmed that ML models using radiomics features extracted from PET images can be useful for predicting late metastasis in patients with early-stage OSCC.