林 邦好

Abstract Recently, artificial intelligence methods have been applied in several fields, and their usefulness is attracting attention. These methods are techniques that correspond to models using batch and online processes. Because of advances in computational power, as represented by parallel computing, online techniques with several tuning parameters are widely accepted and demonstrate good results. Neural networks are representative online models for prediction and discrimination. Many online methods require large training data to attain sufficient convergence. Thus, online models may not converge effectively for low and noisy training datasets. For such cases, to realize effective learning convergence in online models, we introduce statistical insights into an existing method to set the initial weights of deep convolutional neural networks. Using an optimal similarity and resampling method, we proposed an initial weight configuration approach for neural networks. For a practice example, identification of biliary atresia (a rare disease), we verified the usefulness of the proposed method by comparing existing methods that also set initial weights of neural networks.

BACKGROUND: Non-disappearing subsolid nodules requiring follow-up are often detected during lung cancer screening, but changes in their invasiveness can be overlooked owing to slow growth. We aimed to develop a method for automatic identification of invasive tumors among subsolid nodules during multiple health check-ups using radiomics technology based on low-dose computed tomography (LD-CT) and examine its effectiveness. METHODS: We examined patients who underwent LD-CT screening from 2014 to 2019 and had lung adenocarcinomas resected after 5-year follow-ups. They were categorized into the invasive or less-invasive group; the annual growth/change rate (Δ) of the nodule voxel histogram using three-dimensional computed tomography (e.g., tumor volume, solid volume percentage, mean computed tomography value, variance, kurtosis, skewness, and entropy) was assessed. A discriminant model was designed through multivariate regression analysis with internal validation to compare its efficacy with that of a volume doubling time of <400 days. RESULTS: The study included 47 tumors (23 invasive, 24 less invasive), with no significant difference in the initial tumor volumes. Δskewness was identified as an independent predictor of invasiveness (adjusted odds ratio, 0.021; p=0.043), and when combined with Δvariance, it yielded high accuracy in detecting invasive lesions (88% true-positive, 80% false-positive). The detection model indicated surgery 2 years earlier than the volume doubling time, maintaining accuracy (median 3 years vs. 1 year before actual surgery, p=0.011). CONCLUSION: LD-CT radiomics showed promising potential in ensuring timely detection and monitoring of subsolid nodules that warrant follow-up over time.

Studies investigating the associations of oral function and dysphagia with frailty and sarcopenia in community-dwelling older adults are increasing; however, they have not been systematically summarized. We conducted a systematic review to investigate these associations. We searched electronic databases and synthesized relevant data using conventional (frequentist-style) and Bayesian meta-analyses. Twenty-four studies were found to be eligible for our review, including 20 cross-sectional and four prospective cohort studies. Older adults with frailty or sarcopenia had lower tongue pressure, according to the results of conventional meta-analysis (mean difference [95% confidence interval or credible interval]: -6.80 kPa [-10.22 to -3.38] for frailty and -5.40 kPa [-6.62 to -4.17] for sarcopenia) and Bayesian meta-analysis (-6.90 kPa [-9.0 to -4.8] for frailty, -5.35 kPa [-6.78 to -3.89] for sarcopenia). People with frailty had a higher odds ratio (OR) for dysphagia according to the results of conventional meta-analysis (3.99 [2.17 to 7.32]) and Bayesian meta-analysis (1.38 [0.77 to 1.98]). However, the results were inconclusive for people with sarcopenia. A prospective association could not be determined because of the lack of information and the limited number of studies. Decreased oral function and dysphagia can be important characteristics of frailty and sarcopenia in community-dwelling older adults.

OBJECTIVE: Early-stage lung adenocarcinomas that are suitable for limited resection to preserve lung function are difficult to identify. Using a radiomics approach, we investigated the efficiency of voxel-based histogram analysis of 3-dimensional computed tomography images for detecting less-invasive lesions suitable for sublobar resection. METHODS: We retrospectively reviewed the medical records of 197 patients with pathological stage 0 or IA adenocarcinomas who underwent lung resection for primary lung cancer at our institution between January 2014 and June 2018. The lesions were categorized as either less invasive or invasive. We evaluated tumor volumes, solid volume percentages, mean computed tomography values, and variance, kurtosis, skewness, and entropy levels. We analyzed the relationships between these variables and pathologically less-invasive lesions and designed an optimal model for detecting less-invasive adenocarcinomas. RESULTS: Univariate analysis revealed seven variables that differed significantly between less invasive (n = 71) and invasive (n = 141) lesions. A multivariate analysis revealed odds ratios for tumor volumes (0.64; 95% confidence interval (CI), 0.46-0.89; P = .008), solid volume percentages (0.96; 95% CI, 0.93-0.99; P = .024), skewness (3.45; 95% CI, 1.38-8.65; P = .008), and entropy levels (0.21; 95% CI, 0.07-0.58; P = .003). The area under the receiver operating characteristic curve was 0.90 (95% CI, 0.85-0.94) for the optimal model containing these 4 variables, with 85% sensitivity and 79% specificity. CONCLUSIONS: Voxel-based histogram analysis of 3-dimensional computed tomography images accurately detected early-stage lung adenocarcinomas suitable for sublobar resection.

BACKGROUND: Despite multi-professional collaboration via consultation being increasingly important given the variety of disease diagnoses and treatment, the key elements as consultants remain unclear. The study aimed to identify the skills and attitudes that are important for consultants from the residents' perspective so that they can be targeted as priority goals in subsequent educational interventions. METHODS: We conducted our research in two phases: a preliminary survey (May 1 to 14, 2020) and a main survey (June 1 to 14, 2020). As a preliminary survey, first-year postgraduate residents at St. Luke's International Hospital in Tokyo, Japan, were first asked an open-ended question about the types of skills and attitudes that are important for consultants. After eliminating duplicate answers, there were 19 skills and attitudes in total. In the main survey with residents who completed their residency training at our institute, from 2014 to 2018 and current residents (2019-2020), we first asked them about their demographic characteristics (gender, years of postgraduate education, and type of specialty). Then, they answered how important each skill and attitude are for consultants. All 19 items were scored on a seven-point Likert scale that ranged from 0 (completely disagree) to 6 (totally agree). Cronbach's alpha confirmed the internal consistency of the questionnaire items. Principal component analysis and exploratory factor analysis were performed. RESULTS: The survey included 107 individuals (61.1 %, 175 potential participants). The median postgraduate years of education was four (interquartile range: 2-5), and 64.5 % were men (n = 69). Seven key elements for consultants were identified and termed Willing CONSULT. These included (1) willingness (willingness to accept consultation requests), (2) contact (easy access to consultants), (3) needs (consideration of consulters' needs), (4) suggestions and support (providing clear recommendations and suggestions, following up on the patients, and supporting the consulters continuously), (5) urgency (considering the situation's urgency and responding appropriately), (6) learning opportunities (providing teaching points), and (7) text (writing medical records). CONCLUSIONS: We propose Willing CONSULT, which are important skills and attitudes for consultants.

Background: Staphylococcus aureus (SA) is the most common causative microorganism in native vertebral osteomyelitis (NVO). Few studies have compared the clinical features of NVO due to SA (SA-NVO) and NVO due to other organisms (NSA-NVO). This study was conducted to validate a predictive score for SA-NVO to facilitate NVO treatment without broad-spectrum antimicrobial agents. Methods: This retrospective study compared the clinical features of patients with SA-NVO and NSA-NVO who were diagnosed from 2004 to 2019. Univariate associations were assessed using χ 2, Fisher's exact, or Mann-Whitney U test. Multivariable analysis was conducted using logistic regression. The optimal age cutoff point was determined by classification and regression tree analysis. Results: Among 155 NVO patients, 98 (63.2%) had a microbiologically confirmed diagnosis: 40 (25.8%) with SA-NVO and 58 (37.4%) with NSA-NVO. Six predictors, either independently associated with SA-NVO or clinically relevant, were used to develop the STAPH prediction score: atopic dermatitis (Skin) (3 points); recent Trauma (2 points); Age < 67 years (1 point); Abscess (1 point); central venous Port catheter (2 points); and History of puncture (2 points). In a receiver operating characteristic analysis, the area under the curve was 0.84 (95% confidence interval, 0.76-0.91). The best cutoff point was 3. A score ≥3 had a sensitivity, specificity, positive predictive value, and negative predictive value of 58%, 84%, 84%, and 73%, respectively. Conclusions: The STAPH score has relatively high specificity for use by clinicians to predict SA as the causative microorganism in patients with NVO until results of a confirmatory culture are available.

Some studies have shown that sedative antihistamines prolong febrile seizure duration. Although the collective evidence is still mixed, the Japanese Society of Child Neurology released guidelines in 2015 that contraindicated the use of sedative antihistamines in patients with febrile seizure. Focused on addressing limitations of previous studies, we conducted a cross-sectional study to evaluate the relationship between febrile seizure duration and the use of sedative antihistamines. Data were collected from patients who visited St. Luke's International Hospital due to febrile seizure between August 2013 and February 2016. Patients were divided into groups based on their prescribed medications: sedative antihistamine, nonsedative antihistamine, and no antihistamine. Seizure duration was the primary outcome and was examined using multivariate analyses. Of the 426 patients included, sedative antihistamines were administered to 24 patients. The median seizure duration was approximately 3 minutes in all three groups. There was no statistical difference in the bivariate (p = 0.422) or multivariate analyses (p = 0.544). Our results do not support the relationship between sedative antihistamine use and prolonged duration of febrile seizure. These results suggest that the use of antihistamines may be considered for patients with past history of febrile seizure, when appropriate.

BACKGROUND: The stool color card has been the primary tool for identifying acholic stools in infants with biliary atresia (BA), in several countries. However, BA stools are not always acholic, as obliteration of the bile duct occurs gradually. This study aims to introduce Baby Poop (Baby unchi in Japanese), a free iPhone application, employing a detection algorithm to capture subtle differences in colors, even with non-acholic BA stools. METHODS: The application is designed for use by caregivers of infants aged approximately 2 weeks-1 month. Baseline analysis to determine optimal color parameters predicting BA stools was performed using logistic regression (n = 50). Pattern recognition and machine learning processes were performed using 30 BA and 34 non-BA images. Additional 5 BA and 35 non-BA pictures were used to test accuracy. RESULTS: Hue, saturation, and value (HSV) were the preferred parameter for BA stool identification. A sensitivity and specificity were 100% (95% confidence interval 0.48-1.00 and 0.90-1.00, respectively) even among a collection of visually non-acholic, i.e., pigmented BA stools and relatively pale-colored non-BA stools. CONCLUSIONS: Results suggest that an iPhone mobile application integrated with a detection algorithm is an effective and convenient modality for early detection of BA, and potentially for other related diseases.

医師は患者の状態から病気や副作用を予測し,その後の治療計画を立てる.その治療計画に数理科学者の知見を役立たせる余地はあり,特にモデル評価基準に基づく統計的予測手法は重要な役割を果たし得ることが考えられる.本論文では,胸部大動脈瘤に対するステントグラフト治療に焦点を当て,治療後に生じる重篤な副作用であるエンドリーク形成と治療前の胸部大動脈の形状との関係を統計的判別分析を通じて評価する.さらに,各群の判別スコアの平均に対する標本影響関数の符号に基づき,それぞれの群の判別スコアの平均に対して影響的な複数個の学習標本集合を検出する.また,各群の判別スコアの平均に対する標本影響関数の符号に基づき,leave-one-out cross-validationを用いて判別方法の誤判別率を評価し,各群の誤判別率に対して影響的な複数個の学習標本集合についても同様に検出する.最終的に,両者の立場から検出される共通の標本集合の特徴を詳細に調べることにより,エンドリーク形成と患者の大動脈形状との関係を予測の観点から明らかにする.

In this study, we applied a statistical method to quantitative assessment of thoracic aortic morphology. That is, using a discriminant model which predicts a risk of side effect called endoleak after thoracic endovascular aortic repair. Our discriminant model created by a curvature index of aorta and loci of aortic aneurysm predicted the risk of endoleak with 95.5% sensitivity and 87.0% specificity. In addition, the accuracies in the results based on leave-one-out cross-validation were 95.5% sensitivity and 82.6% specificity. Moreover, another model created by a torsional index of aorta as a quantitative information predicted the risk of endoleak with 95.5% sensitivity and 87.0% specificity, and the results of cross-validation were 90.1% sensitivity and 87.0% specificity, respectively. These results led us to the conclusion that the particular parts on the thoracic aorta and the loci of aortic aneurysm strongly relate to a risk of endoleak formation.

Sensitivity analysis based on influence functions has been widely studied in the field of statistics. In particular the evaluation approach has been applied to different statistical methods such as principal component analysis, correspondence analysis, and linear discriminant analysis. However, the study of discriminant methods in pattern recognition is less advanced. With this background, we focused on a subspace method, which is a discriminant method in pattern recognition, and proposed an evaluation method for the influence of training samples to the result of analysis using influence functions. However, the performance and effectiveness of our method were not illustrated well. In this study, we focused on our single-case diagnostics and applied the approach to a representative subspace method, following which we showed good results. Specifically, in situations that had mislabeled samples in the training data, we were able to detect such samples using our approach and subsequently deleted them from the training data to enhance the performance of the target classifier.

In this paper, we discuss sensitivity analysis in linear subspace method, especially on multiple-case diagnostics. Linear subspace method by Watanabe (1973) is a useful discriminant method in the field of pattern recognition. We have proposed its sensitivity analyses, with single-case diagnostics and multiple-case diagnostics with PCA. We propose a modified multiple-case diagnostics using clustering and discuss its effectiveness with numerical simulations.