A new strategy on Early diagnosis of cognitive impairment via novel cross-lingual language markers: a non-invasive description and AI analysis for the cookie theft picture

BackgroundCognitive impairment (CI), including Alzheimers disease (AD) and mild cognitive impairment (MCI), has been a major research focus for early diagnosis. Both speech assessment and artificial intelligence (AI) have started to be applied in this field, but faces challenges with limited language type assessment and ethical concerns due to the "black box" nature. Here, we explore a new stragety with patient led non-invasive observation for a novel cross-lingual digital language marker with both diagnostic accuracy, scalability and interpretability. MethodsSpeech data was recorded from the cookie theft task in 3 cohorts. And automatic speech recognition (ASR), Networkx package, jieba library and other tools were used to extract visual, acoustic and language features. The SHAP model was used to screen features. Logistic regression and support vector machine and other methods were used to build the model, and an independent cohort was used for external verification. Finally, we used AIGC technology to further reproduce the entire task process. ResultsIn Chinese environment, we built 3 models of NC/aMCI, NC/AD, and NC/CI (aMCI+AD) through Cohort 1 (NC n=57, aMCI n=62, AD n=66), with accuracy rates of 0.83, 0.79, and 0.79 respectively. The accuracy was 0.75 in the external scalability verification of Cohort 3 (NC n=38, CI n=62). Finally, we built a cross-lingual (Chinese and English) model through Cohort 1 and 2, built a NC/aMCI diagnosis model, and the diagnostic accuracy rate was 0.76. Lastly, we successfully recreate the testing process through Text-to-Image and Animation Generation. DiscussionThe visual features created by our research group and combines acoustic and linguistic features were used to build a model for early diagnosis of cognitive impairment, and a cross-lingual model covering English and Chinese, which performs well in external verification of independent cohorts. Finally, we innovatively used AI-generated videos to show the subjects task process to the physician to assist in judging the patients diagnosis. Keyword: Alzheimers disease, Amnestic mild cognitive impairment, speech test, Artificial Intelligence, interpretability