市野 直浩

OBJECTIVE: Epilepsy treatment with anti-seizure medications (ASMs) is based on careful assessment of the balance between the likelihood of further seizures and the risk of side effects of treatment. However, there is currently no established biomarker to ascertain seizure control status with ASMs. High-frequency oscillations (HFOs), transient bursts of EEG activity with frequencies beyond 80 Hz, are a new and promising noninvasive epilepsy biomarker. We compared the risk of scalp HFO appearance between pediatric patients with good and poor seizure control by treatment with ASMs. METHODS: A total of 72 epilepsy patients (aged 0-18 years, 39 males) with good and poor seizure control with ASMs participated in this study. We applied a validated automated detector to determine HFO and spike. We calculated the odds ratios (ORs) for scalp HFO and spike appearance according to seizure control status by multiple logistic regression analysis. RESULTS: Scalp HFO was seen more commonly and with a significantly higher detection rate in patients with poor seizure control as compared with patients with good seizure control for both ripple and fast ripple. These significant associations were found for both focal and generalized epilepsy. The ORs for scalp HFO appearance adjusted for confounding factors were significantly higher in patients with poor seizure control compared to those with good seizure control (ripple: OR [95% CI] = 11.91 [2.21-64.30], p = 0.004; fast ripple: 4.98 [1.03-24.09], p = 0.046). There were no significant associations between spike appearance and seizure control status. SIGNIFICANCE: We found an increased risk of scalp HFO appearance in patients with poor seizure control. The results of this study support that scalp HFO is associated with patients having frequent seizures after treatment in both ripple and fast ripple. PLAIN LANGUAGE SUMMARY: This study analyzed scalp high-frequency oscillations and spikes in pediatric patients with various types of epilepsy who were being treated using ASMs. The results showed that an increased risk of scalp HFO appearance was observed in patients with poor seizure control compared to those with good seizure control. These findings were observed in both the ripple (80-250 Hz) and fast ripple (250-500 Hz) bands. The scalp HFO is associated with patients having frequent seizures after treatment in both ripple and fast ripple.

近年,大規模言語モデル(large language models;LLM)が世界的に様々な分野で注目を集めている。LLMとは,非常に巨大なデータセットとディープラーニング技術を用いて構築された言語モデルである。LLMは,人間に近い流暢な会話が可能であら,自然言語を用いたさまざまな処理を高精度で行えることから,世界中で注目を集めている。本研究では,LLMであるOpenAI社が開発したChatGPTの異なる2つのモデル(GPT-3.5,GPT-4)にて,過去3年間の臨床検査技師国家試験におけるChatGPTの正答率について評価を行った。GPT-3.5による正答率の平均は51.4%であった。一方,GPT-4では79.8%の正答率結果が得られた。本結果より,ChatGPTはこの先医療現場における有効なアドバイザーとして進化する可能性をもつことが示唆された。しかし,今回不正解となった20%の中には,患者を診断する際に誤診につながりかねない回答が含まれており,今後のChatGPTの精度向上は必須と考えられる。今回の検証は,LLMにおけるChatGPTの臨床検査領域での多様な応用の進展に寄与すると考えられ,この先の発展に期待したい。(著者抄録)