【作者】 张统一； 赵鑫；

【Author】 ZHANG TongYi;ZHAO Xin;Key Laboratory of Behavioral and Mental Health of Gansu province;School of Psychology,Northwest Normal University;

【机构】 甘肃省行为与心理健康重点实验室； 西北师范大学心理学院；

【摘要】 执行功能已被广泛视为解码精神分裂症病理机制的关键途径,在精神分裂症的发病机制和病程中起重要作用。然而,由于受传统统计方法的限制,目前对执行功能与精神分裂症之间复杂关系的理解仍十分有限。传统统计方法并不能充分捕捉到执行功能特征与精神分裂症之间的隐含联系与内在规律。本研究首次尝试使用可解释性机器学习算法,以揭示执行功能在精神分裂症诊断和预后中的潜在价值。实验一基于195名精神分裂症患者和169名健康对照组的执行功能行为数据和社会人口学资料,使用Stacking方法融合三种基学习器(SVM、Adboost以及ANN模型)建模,并采用Optuna算法参数调优,之后利用SHAP值进行模型解释。实验二构建纳入与不纳入执行功能行为数据的两组模型(两组模型均整合临床易获取的数据,如临床量表、病历、血常规等),比较模型在预测86名患者住院治疗4至6周后临床症状缓解效果的性能。结果显示,OP-Stacking (Optuna-Stacking)模型在分类诊断任务中表现最佳(准确率=0.8783;精确度=0.8710;召回率=0.8896;F1值=0.8774;AUC=0.9359)。流体智力、失业、独生子女以及执行功能被确认为模型最重要的预测因素。在基于执行功能特征预测个体是否为精神分裂症患者时,模型表现出一定的差异性。在预后分类任务中,纳入执行功能行为数据的模型在预测精神分裂症治疗效果方面的性能更优。结果发现,执行功能可作为一种可靠的生物标志物,用于精神分裂症的辅助诊断和预后评估。基于执行功能行为数据和易获取的社会人口学信息所构建的分类模型具有较好分类性能,能有效识别出精神分裂症患者。尽管执行功能损伤是精神分裂症群体的典型表现,但可能并非所有亚型的患者都在同等程度上受到影响。临床诊断时应关注那些执行功能损伤的独生子女以及大龄失业青年。执行功能在预测精神分裂症临床干预效果方面也起到了关键作用。未来研究应进一步探索执行功能在诊断和预后评估共病或类似精神疾病患者中的作用,尤其是关注执行功能三个子成分(转换、抑制与刷新)同一性和独特性的潜在价值。更多还原

【Abstract】 Executive functions have been widely recognized as pivotal pathways for decoding the pathological mechanism of schizophrenia,playing an important role in the disease’s onset and progression.However,due to the constraints of traditional statistical methods,our current understanding of the intricate relationship between executive functions and schizophrenia remains limited.Traditional statistical methods fail to fully capture the implicit connections and intrinsic patterns between the characteristics of executive functions and schizophrenia.This study is the first to attempt to leverage interpretable machine learning algorithms to reveal the potential of executive functions in diagnosing and predicting schizophrenia.In Experiment 1,we utilized behavioral data related to executive functions and sociodemographic information from 195 schizophrenia patients and 169 healthy controls.We used a Stacking methodology to combine three base learners(Support Vector Machine,Adaboost,and Artificial Neural Network),and the Optuna algorithm was employed for parameter tuning.Subsequently,we used SHAP values for model interpretation.In Experiment 2,we constructed two groups of models,one incorporating executive function behavioral data and the other excluding it(both sets of models integrated readily obtainable clinical data,such as clinical scales,medical records,and routine blood tests),and compared their performance in predicting the remission of clinical symptoms in 86 patients after4 to 6 weeks of hospital treatment.The results revealed that the Optuna-Stacking(OP-Stacking) model performed best in the diagnostic classification task(accuracy=0.8783;precision=0.8710;recall=0.8896;F1 score=0.8774;AUC=0.9359).Fluid intelligence,unemployment,being an only child,and executive functions were confirmed as the model’s most important prediction factors.The model showed some variability when predicting whether an individual is a schizophrenia patient based on the characteristics of executive functions.In prognosis classification tasks,the model that incorporated executive function behavioral data demonstrated superior performance in predicting the treatment outcomes of schizophrenia.These results indicate that executive functions can serve as reliable biomarkers for assisting in the diagnosis and prognostic assessment of schizophrenia.The classification model,built on the basis of executive function behavioral data and easily obtainable sociodemographic information,demonstrated great classification performance and could effectively identify schizophrenia patients.Although impairment in executive functions is a typical manifestation in the schizophrenia population,it may not affect all subtypes of patients to the same extent.Clinical diagnoses should pay attention to single children and older unemployed individuals with impairments in executive functions.Executive functions also played a crucial role in predicting the effectiveness of clinical interventions for schizophrenia.Future research should further explore the role of executive functions in the diagnosis and prognostic evaluation of patients with comorbidity or similar mental disorders,especially focusing on the potential of the unity and diversity of the three subcomponents of executive functions(shifting,inhibition,and updating).更多还原