基于机器学习动态简化的明尼苏达多相个性调查表  

Minnesota Multiphasic Personality Inventory based on machine learning dynamic simplification

作　　者：孙启科 董问天[2] 迟锐 贺勇 马义斌 王克[3] 崔飞环 冯超南 李毓明 于淏岿 于滨 石川[2] 纪俊[1,3] SUN Qike;DONG Wentian;CHI Rui(Qingdao University, Qingdao 266071, China)

机构地区：[1]青岛大学,山东青岛266071 [2]北京大学第六医院,北京大学精神卫生研究所,国家卫生健康委员会精神卫生学重点实验室(北京大学),国家精神心理疾病临床医学研究中心(北京大学第六医院) [3]青岛市市立医院 [4]首都医科大学附属北京安贞医院 [5]北京万灵盘古科技有限公司

出　　处：《精神医学杂志》2021年第2期113-118,共6页Journal of Psychiatry

基　　金：国家自然科学基金项目(编号:61503208);山东省自然科学基金项目(编号:ZR2015PF002)。

摘　　要：目的使用机器学习算法对明尼苏达多相个性调查表(MMPI)中的临床量表进行简化。方法将7410例测试者的测评数据针对不同人群进行划分,通过机器学习算法进行建模分析以实现动态删减部分题组。结果将动态删减后的结果保持删减前结果75%的敏感度与特异度为阈值,男性的癔症(Hy)为阴性且精神衰弱(Pt)、精神分裂(Sc)为阳性的分组,癔症(Hy)、抑郁(D)为阴性且精神衰弱(Pt)、精神分裂(Sc)为阳性的分组,精神分裂(Sc)、精神衰弱(Pt)、妄想(Pa)、男性-女性倾向(Mf)、病态人格(Pd)、社会内向(Si)为阴性且癔症(Hy)、抑郁(D)为阳性的分组,均可缩减8.3%的题目;抑郁(D)、精神衰弱(Pt)、精神分裂(Sc)、社会内向(Si)、轻躁狂(Ma)、男性-女性倾向(Mf)为阴性且疑病(Hs)为阳性的分组可缩减15.0%的题目;精神分裂(Sc)为阴性且抑郁(D)为阳性的分组可缩减12.0%的题目。女性的精神分裂(Sc)、抑郁(D)、病态人格(Pd)、社会内向(Si)、疑病(Hs)、妄想(Pa)、轻躁狂(Ma)、男性-女性倾向(Mf)为阴性且癔症(Hy)为阳性的分组可缩减12.0%的题目;抑郁(D)、社会内向(Si)、癔症(Hy)为阴性且精神分裂(Sc)、妄想(Pa)、精神衰弱(Pt)为阳性的分组可缩减11.5%的题目。结论简化后的版本实现了筛查与辅助诊断的功能,更适用于体检场景。基于本研究提出的机器学习模型,继续扩大数据量,能够挖掘出更多的可简化分组。Objective To simplify the clinical scale in Minnesota Multiphasic Personality Inventory(MMPI)through machine learning algorithm.Methods Evaluation data from 7410 subjects were divided into different groups,and machine learning algorithm was used for modeling analysis to achieve dynamic deletion of some item groups.Results 75%of the sensitivity and specificity of the results before deletion were maintained as the threshold for the results after dynamic simplification.The group of males whose Hy was negative and Pt,Sc were positive could reduce 8.3%of items.The group of males whose Hy,D were negative and Pt,Sc were positive could reduce 8.3%of items.The group of males whose Sc,Pt,Pa,Mf,Pd,Si were negative and Hy,D were positive could reduce 8.3%of items.The group of males whose D,Pt,Sc,Si,Ma,Mf were negative and Hs was positive could reduce 15.0%of items.The group of males whose Sc was negative and D was positive could reduce 12.0%of items.The group of females whose Sc,D,Pd,Si,Hs,Pa,Ma,Mf were negative and Hy was positive could reduce 12.0%of items.The group of females whose D,Si,Hy were negative and Sc,Pa,Pt were positive could reduce 11.5%of items.Conclusion The simplified version realizes the functions of screening and auxiliary diagnosis,and is more suitable for physical examination scenarios.Based on machine learning model proposed in this study,as the amount of data increases,more simplified groups can be explored.

关 键 词：MMPI 机器学习 动态简化 

分 类 号：B841[哲学宗教—基础心理学]