KKKenji Kondo

Kenji Kondo

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Kenji Kondo has a total of 9 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Benchmarking Local Large Language Models for Predicting Vestibular Dysfunction Based on Caloric Testing: A Comparison with Conventional Machine Learning on Stabilometry Data

    by Teru Kamogashira, Kento Koda, Kentaro Ichijo, Mineko Oka, Kenji Kondo, Makoto Kinoshita, Chisato Fujimoto

    Background/Objectives: In-context learning enables large language models (LLMs) to perform clinical prediction without task-specific training; however, their performance on structured tabular data relative to conventional machine learning remains unclear. This exploratory study evaluated the performance, scalability, and efficiency of local LLMs for predicting caloric-test-defined vestibular dysfunction from stabilometry data. Methods: We analyzed stabilometric data from 238 participants (75 with vestibular dysfunction, 163 controls). Forty-three local LLMs spanning multiple families were evaluated using in-context learning on a local setup using CPU-dominant inference with partial GPU offloading, without external data transmission. Performance and inference time were measured via repeated hold-out validation and compared with logistic regression (LR) and tree-based ensemble methods. Results: The tree-based ensemble models showed the highest predictive performance. LLM performance was heterogeneous and varied by model family and size. Mid-sized models achieved accuracy comparable to LR, although their performance was not statistically distinguishable from either LR or the majority-class baseline, whereas smaller models frequently failed to produce usable binary outputs. Larger models did not consistently provide proportional gains relative to their computational cost. Conclusions: Local LLMs did not outperform conventional machine learning for maximizing predictive accuracy on this small tabular dataset. However, they may serve as a complementary, privacy-preserving approach that enables deployment without task-specific training. Future work should validate these findings with larger datasets, more robust validation, standardized outcome definitions, and constrained decoding.

  • Prognostic Impact of the 10% Electroneurography Threshold on Complete Recovery in Pediatric Bell’s Palsy: A Hazard Ratio Analysis

    by Shintaro Baba, Ai Yoshitomi, Ryosuke Sugito, Kenji Kondo

    Objective: To investigate the ability of the 10% electroneurography (ENoG) threshold to predict complete recovery (House-Brackmann [H-B] grade I without sequelae) in pediatric Bell’s palsy. Methods: This retrospective study included 27 pediatric patients aged 0 to 12 years with Bell’s palsy who underwent ENoG between 8 days and 3 weeks after onset. All the patients received corticosteroid therapy (prednisolone ≥1 mg/kg/day), with severe cases receiving a high-dose regimen (2–4 mg/kg/day). The primary endpoint was complete recovery, defined as H-B grade I with no pathologic synkinesis or facial contracture. Recovery profiles were compared between a high-ENoG group (ENoG ≥10% residual amplitude, n=13) and a low-ENoG group (ENoG P P <0.001). Conclusion: While pediatric Bell’s palsy had an excellent overall prognosis in terms of functional improvement (H-B grade II), the ENoG 10% threshold was the best predictor of complete recovery without permanent sequelae.