Fig 1.
a) An overview of our methodology. The coherence between EEG electrode pairs in different frequency bands and the modified UPSIT (Iran-SIT) score are used as features for training an SVM classifier. The selection of the odorant subset in the UPSIT kit and the significant frequency bands and connections in the EEG records is carried out by statistical analysis. b) EEG response to the sequence of stimuli is pre-processed and epochs are extracted for further analysis. c) A measure of coherence deficit is calculated between each pair of channels for the gamma oscillation band. Similar operations are performed for each of the other oscillatory bands (delta, theta, alpha, beta).
Table 1.
Participant characteristics: P-values denote the separation between healthy participants and mild AD patients in each characteristic.
Table 2.
MMSE results.
Fig 2.
Total and modified UPSIT scores plotted versus age: a) Top: Regression of total UPSIT score with age. Distributions of total UPSIT scores for the two groups are shown as histograms. Bottom: Effect of age on total UPSIT score is removed and residual values of the two groups are histogrammed. b) Top: Regression of modified UPSIT score with age. Distributions of modified UPSIT scores for the two groups are shown as histograms. Bottom: Effect of age on modified UPSIT score is removed and residual values of the two groups are histogrammed. The 95% confidence intervals are also plotted in the top figures.
Table 3.
P-values derived from t-test for significant UPSIT odors separating the two groups of participants.
Fig 3.
a) Percentage of correct answers for each UPSIT (Iran-SIT) odor identification question. The x-axis denotes the tested odors and is sorted from the left by the number of correct answers that the mild AD patients gave. Hence, the leftmost odor is the one that the mild AD patients identified most. The y-axis is the percentage of correct answers to each odor identification question in each group. Two odors that are statistically significant in distinguishing between mild AD patients and healthy participants are denoted by + and x marks. b) The number of correct answers for each UPSIT (Iran-SIT) odor identification question divided into five-year age bins. The shade of each bin denotes the number of correct answers. Green (light) pixels indicate that most participants in the corresponding age bin answered the question correctly, and the blue (dark) pixels suggest that most of the participants were unable to identify the presented odor. The upper diagram is for the healthy participants, and the lower diagram is for the mild AD patients. The two statistically significant odors are denoted by dashed boxes.
Fig 4.
a) Beta-band component of a sample epoch from the data of the Fz and Cz electrodes for one AD (top) and one healthy (bottom) participant. b) Gamma-band component of a sample epoch from the data of the Fz and Cz electrodes for one AD (top) and one healthy (bottom) participant. c) Top: Regression of beta-band ImCoh value with age. Values for the two groups are shown as histograms. Bottom: Age-adjusted beta-band ImCoh values are shown as histograms for the two groups. d) Top: Regression of gamma-band ImCoh value with age. Values for the two groups are shown as histograms. Bottom: Age-adjusted gamma-band ImCoh values are shown as histograms for the two groups. e) Statistical significance of the gamma-band ImCoh value difference between mild AD patients and healthy participants. The Fz-Cz connection is the most significant connection with p-value < 0.05. f) The ImCoh value of the Fz-Cz connection measured in each frequency band. Statistically significant frequency bands are denoted by + and x marks.
Table 4.
Classification accuracy for different modalities and the multi-modal analysis based on significant components of each modality.
a) Feature data were used for classification disregarding the age of the participants. b) Residuals of each feature relative to its corresponding healthy group regressor were used for classification.
Fig 5.
a) Correlation of the total UPSIT score and the MMSE score. b) Correlation of the beta-band ImCoh value and the MMSE score. c) Correlation of the gamma-band ImCoh value and the MMSE score.
Fig 6.
Deposition of amyloid-β plaques in the frontal lobe leads to the weakening of neuronal connections.
Neuronal populations are still working, but they are not synchronous enough to send effective inhibitory feedback to earlier processing layers.