Tetsuo Ishikawa

Background: Anemia is a common systemic condition associated with adverse maternal, perioperative, and cardiovascular outcomes. Although timely screening is clinically important, diagnosis still relies on invasive blood testing. Palpebral conjunctival pallor has traditionally been used as a noninvasive indicator of anemia, but its diagnostic accuracy remains limited. This study aimed to develop and validate a deep learning system to estimate hemoglobin (Hb) concentration and screen for anemia using palpebral conjunctiva images captured with a smartphone-compatible slit-lamp microscope. Methods: In this prospective observational study, 225 Japanese participants (20–92 years) underwent conjunctival imaging and blood testing. Palpebral conjunctiva videos were obtained using the Smart Eye Camera. Video frames were processed using automated anterior-segment segmentation and conjunctiva extraction. A ConvNeXt-based regression model was trained to predict Hb values. Anemia was defined using sex-specific Hb thresholds. Results: From 225 videos, 53,776 frames were extracted, yielding 9903 quality-filtered conjunctiva images (training: 8082; test: 1821). Video-level predicted Hb values moderately correlated with measured Hb (r = 0.42). For anemia screening, frame-level analysis achieved an AUC of 0.75, with accuracy of 0.76, sensitivity of 0.71, and specificity of 0.79. Video-level aggregation achieved 69% accuracy. Conclusions: Deep learning analysis of palpebral conjunctiva images acquired with a portable slit-lamp microscope demonstrated the feasibility of non-invasive hemoglobin estimation and anemia screening. Although the proposed approach achieved moderate performance, further improvements in model accuracy and prospective multi-center validation are required before clinical implementation.

Background: Recurrence after surgery for chronic subdural haematoma (CSDH) is common, yet reported predictors are inconsistent and it is unclear how accurately recurrence can be predicted. We quantified achievable preoperative accuracy and derived a pragmatic risk score. Methods: In this multicentre retrospective study at 15 Japanese centres (2020–2022), we analysed consecutive adults undergoing burr-hole drainage for CSDH; the primary outcome was ipsilateral reoperation for recurrence. Using 58 preoperative variables and five imputed datasets, we estimated discrimination by leakage-free nested cross-validation across eight machine-learning algorithms, and derived an integer score assessed by calibration, decision-curve analysis and internal–external cross-validation (leave-one-institution-out). Findings: Of 1159 surgical records, 860 patients were analysed; 80 (9.3%) recurred. No algorithm exceeded a c-statistic of 0.70 (best 0.60, 95% CI 0.54–0.67), and predictive information was diffuse, non-additive and context-dependent. A four-item score (male sex, BMI < 19 kg/m², APTT ≥ 31 s, high-risk CT density) showed graded recurrence (1.5–26.3%) and an apparent AUC of 0.68 (0.66 after cut-point correction). Under internal–external cross-validation the pooled c-statistic was 0.67 (95% CI 0.59–0.73), with reproducible discrimination but centre-dependent calibration. Interpretation: Under leakage-free validation, no algorithm exceeded a c-statistic of 0.70: preoperative prediction of CSDH recurrence has a consistent, possibly power-limited ceiling that earlier, less rigorously validated studies likely overstated. A simple four-item score matched this ceiling and, after local recalibration, can support low-burden, risk-stratified care. Recurrence appears multifactorial, with much of the risk unexplained by routine preoperative variables, pointing to subdural-space biology and peri-operative factors as priorities.

Manifold-based single-cell omics analyses assume that high-dimensional observations can fall into a low-dimensional space encoding biological constraints. In practice, per-cell observation is highly heterogeneous: Shallowly and deeply observed cells coexist. In an empirical single-cell RNA sequencing dataset, shallowly observed cells cluster together to generate spurious hubs that can give rise to illusory loops in low-dimensional manifold skeletons in graph abstraction. Several imputation methods leave these artifacts largely intact, whereas graph abstraction restricted to homogeneously observed cells alone recovers tree-like structures locally representing constrained cell state transitions. Simulations further demonstrate that realistic heterogeneous observation can create spurious subclusters and false branching. Grounded by these observations, we propose topological stability descriptors of low-dimensional manifold skeletons to delineate a regime in which manifold-based inference is trustworthy despite realistic heterogeneous observations. Our findings underscore that the heterogeneity of observations is not merely noise but a source of systemic distortion in manifold-based inference that must be addressed.

Background: The stay-at-home orders, lockdowns, and states of emergency of the Coronavirus Infectious Disease emerged in 2019 (COVID-19) pandemic have affected the mental health of school-aged children. Previous reports of psychological distress in adolescents during the pandemic have been mixed, however, with some reports showing increases in psychological distress and others suggesting decreases. To accurately assess the impact of the pandemic, we need to be able to compare psychological assessments longitudinally, both before and during the pandemic. However, current statistical methods have limitations for reconstructing the complex trajectory of psychological states as captured by short-item questionnaires. Methods and findings: In this study, we analyzed monthly Kessler 6-item Psychological Distress Scale (K6) questionnaire responses collected from 16- to 18-year-old high school students participating in the population-neuroscience Tokyo TEEN Cohort (pn-TTC) in Japan (1,278 responses from 84 participants). Participants included 42 males and 42 females. The pn-TTC is a population-based longitudinal study conducted in Tokyo, Japan that follows children to investigate their developmental and mental health trajectories. In addition to conventional statistical approaches that summarize multiple questionnaire items into a composite score, we applied “energy landscape analysis,” a method derived from statistical physics that models multivariate psychological states as a dynamic system of interactions among K6 questionnaire items, to visualize longitudinal changes in psychological distress before and during the COVID-19 pandemic (July 2019 to September 2021). Here, we define the depressive and healthy states as configurations in which all six K6 items are above or below each participant’s individual mean, respectively. Before the pandemic, the healthy state occurred 11.0 times as frequently as the depressive state. In contrast, during the pandemic, the relative frequency of the healthy state increased to 18.2, 18.5, and 15.0 times that of the depressive state, respectively. The evolving energy landscape revealed an association between the pandemic period and a lower likelihood of being in a depressive state. We also identified two groups of students with different K6 dynamics and energy landscapes. The first group consisted of 61 participants whose total K6 score was relatively low (less than 5) and stable over time, and the second group consisted of 23 participants whose total K6 score was higher (with most being higher than 5) and less stable. The latter group showed a greater change in cortical thickness in the caudal part of the middle frontal gyrus (cMFG) ( t -statistic = −2.36, p -value = 0.019, q -value = 0.048) and the temporal pole (TP) ( t = 3.08, p = 0.0023, q = 0.012), as measured by magnetic resonance imaging, in the direction of accelerated adolescent brain development. Because all participants lived in Tokyo, generalizability remains limited, and as the association between psychological states and brain development is descriptive, future studies in diverse cohorts are needed to examine causality. Conclusions: By revealing associations between the COVID-19 pandemic and lower levels of psychological distress and healthier mental health states, our work demonstrates the potential of using dynamical systems theory, such as the energy landscape analysis, to interpret health and disease metrics in psychology and psychiatry. This approach may improve mental health surveillance for the next pandemic.

Manifold-based single-cell omics analyses assume that high-dimensional observations can fall into a low-dimensional space encoding biological constraints. In practice, per-cell observation is highly heterogeneous: shallowly- and deeply-observed cells coexist. In an empirical scRNA-seq dataset, shallowly-observed cells cluster together to generate spurious hubs that can give rise to illusory loops in low-dimensional manifold skeletons in graph abstraction. Several imputation methods leave these artifacts largely intact, whereas graph abstraction restricted to homogeneously-observed cells alone recovers tree-like structures locally representing constrained cell state transitions. Simulations further demonstrate that realistic heterogeneous observation can create spurious subclusters and false branching. Grounded by these observations, we propose topological stability descriptors of low-dimensional manifold skeletons to delineate a regime in which manifold-based inference is trustworthy despite realistic heterogeneous observations. Our findings underscore that the heterogeneity of observations is not merely noise but a source of systemic distortion in manifold-based inference that must be addressed.

Regular and sufficient sleep is essential for children's development. Sleep deficiency is related to various conditions, such as autism spectrum disorder and attention-deficit hyperactivity disorder. Parent-monitored sleep logs are widely used, but cannot provide comprehensive assessment of paediatric sleep rhythms, because each child's developmental stage, diurnal variations, and daily fluctuations need to be considered. To address this limitation,we developed an artificial intelligence (AI) -based approach that combines unsupervised machine-learning techniques to analyze pediatric sleep data. This approach allowed us to analyse various sleep features comprehensively and visualize typical sleep development. To illustrate the application of our approach, we conducted an association analysis with developmental disorders. We identified specific characteristics of children with developmental disorders. We also introduced a new metric, the degree of deviation from typical development, which provided a more accurate identification of their sleep patterns than achieved with traditional metrics. Our AI-based method may be useful for identifying abnormal sleep rhythm features that could be associated with developmental disorders using practical parent-reported data.

Modern clinical oncology faces an unprecedented data complexity that exceeds human analytical capacity, making artificial intelligence (AI) integration essential rather than optional. This review examines the dual impact of AI on productivity enhancement and creative discovery in cancer care. We trace the evolution from traditional machine learning to deep learning and transformer-based foundation models, analyzing their clinical applications. AI enhances productivity by automating diagnostic tasks, streamlining documentation, and accelerating research workflows across imaging modalities and clinical data processing. More importantly, AI enables creative discovery by integrating multimodal data to identify computational biomarkers, performing unsupervised phenotyping to reveal hidden patient subgroups, and accelerating drug development. Finally, we introduce the FUTURE-AI framework, outlining the essential requirements for translating AI models into clinical practice. This ensures the responsible deployment of AI, which augments rather than replaces clinical judgment, while maintaining patient-centered care.

Background: Patients’ clinical status often evolves rapidly after an initial diagnosis, with each patient exhibiting a distinct disease trajectory. As a result, static risk scores fall short in supporting timely interventions—an issue highlighted by COVID-19, where deaths have stemmed from heterogeneous pathways such as pneumonia, multiorgan failure, or exacerbation of preexisting conditions. Objective: This study aims to propose a dynamic prognostic risk assessment framework based on longitudinal data collected during hospitalization, using COVID-19 as an example. Our aim was to develop and validate an interpretable framework that (1) screens prognosis at admission and (2) dynamically updates mortality risk throughout hospitalization, thereby providing clinicians with early, explainable warnings while minimizing additional cognitive load. Methods: In this retrospective study, we extracted electronic medical records of 382 COVID-19 cases treated at Tokyo Shinagawa Hospital between January 27 and September 30, 2020. At admission, gradient boosting decision trees (Light Gradient Boosting Machine) were used to predict the maximum clinical deterioration, including death, based on data available at initial diagnosis. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). For in-hospital monitoring, random survival forests (RSF) were trained on a longitudinal dataset that combined static demographic characteristics with serially measured vital signs and laboratory results. The model dynamically assessed daily mortality risk by calculating a 7-day cumulative hazard function, with risk scores recalculated each day during hospitalization. RSF accuracy was evaluated in an independent one-third test set using the concordance index (C-index), an integrated Brier score (1-50 days), and mean time-dependent AUC. SurvSHAP(t), an extension of Shapley Additive Explanations, was applied to provide time-dependent explanations of each variable’s contribution to the prediction. Results: The prediction at initial diagnosis showed good agreement with the actual severity outcomes (AUC of 0.717 for predicting hospitalization/severity ≥2; 0.878 for severity ≥3; 0.951 for severity ≥4; 0.952 for severity ≥5; and 0.970 for death/severity=6), although some cases exhibited discrepancies between the predicted and actual prognoses. The dynamic mortality risk assessment during hospitalization using the RSF achieved a test-set C-index of 0.941, an integrated Brier score of 0.315, and a mean time-dependent AUC of 0.936. This dynamic assessment was able to distinguish between dead and surviving patients as early as 1‐2 weeks before the outcome. Early in hospitalization, C-reactive protein was an important risk factor for mortality; during the middle period, peripheral oxygen saturation (SpO₂) gained importance; and immediately before death, platelets and β-D-glucan were the primary risk factors. Conclusions: Integrating static admission triage with daily, explainable RSF predictions enables early identification of patients with COVID-19 at high risk of deterioration. By surfacing phase-specific, actionable predictors, the framework supports timely interventions and more efficient resource allocation. Prospective, multicenter studies are warranted to validate its generalizability and clinical impact.