過去研究顯示使用P300作為指標的隱藏訊息測驗(Concealed Information Test)可以有效鑑別知情與不知情的受試者。本研究旨在評估P300鋒間振幅(P300 peak-to-peak magnitude)用於偵測從模擬犯罪情境中獲得之犯罪訊息的效率。本研究包含兩個實驗。實驗一中受試者被要求執行一項模擬竊盜案，受試者必須假裝偷盜一個重要物品並且將之藏起來。實驗二中受試者被要求執行一項間諜活動，其中兩位受試者彼此會在一個指定地點見面，並且交接一件機密物品。受試者於是接受隱藏訊息腦波測謊來檢測其是否知道犯罪相關細節，而在測謊的過程中腦波會被記錄。接著腦波資料經過前處裡程序，包括0.3-50 Hz的帶通濾波、降低取樣頻率到250 Hz、利用人為子空間重建(Artifact Subspace Reconstruction)及獨立成分分析(Independent Component Analysis)來移除雜訊，再使用6 Hz低通濾波。然後使用自助重抽振幅差異法(Bootstrap Amplitude Difference)來比較受試者看到犯罪相關刺激與犯罪無關刺激時所產生的P300大小，對受試者進行有無罪分類，並計算四道犯罪訊息題目的鑑別率。結果顯示兩實驗的隱藏訊息腦波測謊獲得中等的鑑別率(.62-.81)，代表P300鋒間振幅在本研究中並非一個穩定有效的鑑別特徵。;The Concealed Information Test (CIT) with P300-based EEG examination has been reported of high performance in differentiating knowledgeable/non-knowledgeable subjects. This study evaluated the efficacy of the P300 peak-to-peak (p-p) magnitude in the detection of the guilty knowledge acquired during mock crime scenarios. Two experiments with different mock crime scenarios were conducted. The mock crime in Experiment 1 was a theft crime, in which participants pretended to steal a critical item and hide it to a designated place. The mock crime in Experiment 2 was an espionage scenario, in which two participants met each other at a specific place and exchanged a secret item. Participants received the CIT EEG examination for the test of guilty knowledge during which the EEGs were recorded. EEG was then preprocessed through a band-pass filter with passband in between 0.3-50 Hz, downsampled to 250 Hz, and undergone the Artifact Subspace Reconstruction (ASR) and Independent Component Analysis (ICA) to remove noise. Finally, after a further 6 Hz low-pass filtering, the Bootstrap Amplitude Difference (BAD) was used to compare the P300 peak-to-peak magnitudes of the probe and irrelevant conditions. If the P300 peak-to-peak magnitude of the probe condition was greater than that of the irrelevant condition, it was more likely that the participants should come from the guilty group. Otherwise, the participant might belong to the innocent group. The classification accuracy was then computed channel by channel across all subjects. The results showed only a moderate classification accuracy (.62-.81) and indicated that the P300 p-p magnitude might not be a suitable feature in the present study.
