Table 1.

Overview of the most widely used datasets and comparative benefits of our new proposed multimodal longitudinal dataset

Dataset	Population	Samples	Modality	Longitudinal	Elicitation	Duration	Task
DementiaBank (Becker et al., 1994)	196 Dem vs. 98 Ctrl	255 Dem vs. 244 Ctrl	Audio$†$	1–5 sessions	CTP Description	1.90	AD Class, Score Reg
Carolinas(Pope & Davis, 2011)	125 Dem vs 125 Ctrl	400 Dem Vs 250 Ctrl	Text$+$Audio$+$Video$\star$	1–9 sessions	Health and Well-being Conversations	76.14	AD Class, Detection of Confusion
ADReSS(Luz et al., 2020)	78 Dem vs. 78 Ctrl	78 Dem vs. 78 Ctrl	Text$+$Audio$†$	No	CTP Description	1.20	AD Class, Score Reg
$ADReS{S}_o$(1) (Luz et al., 2021)	32 Dem	105	Audio$‡$	No	Picture Description	2.17	Disease Progression
$ADReS{S}_o$(2) (Luz et al., 2021)	36 Dem vs. 35 Ctrl	237	Audio$‡$	No	Picture Description	2.21	AD Class, Score Reg
Carers’ interactions with dementia patients (Hansebo & Kihlgren, 2002)	14 Dem	14 Dem	Text$+$Video$‡$	No	Video-recorded Interactions	–	Qualitative Analysis
ILSE (Weiner et al., 2016)	23 Dem	112 hours recording	Text$+$Audio$\star$	1-4 sessions	Recorded Interactions	–	Disease Progression
Verbal fluency and brain imaging scores (Clark et al., 2016)	107 Dem vs. 51 Ctrl	-	Text+MRI scores$†$	1-4 sessions	Recorded Interactions	–	Disease Progression
Our data collection	14 Dem vs. 8 Ctrl	408 Dem vs. 408 Ctrl	Audio+Text+Keyboard+Pen	56 sessions	Reminiscence Materials	12.25	Longitudinal Language Changes

Elicitation data elicitation task, Duration average speech duration of the elicited task in minutes, CTP cookie theft picture, Dem participants with dementia, Ctrl participants with no dementia diagnosis, AD Class Alzheimer’s dementia classification. Score Reg score regression

$\star$ Conversations, $†$ Monologue speech, $‡$ Occasional interactions between clinicians and participants