Diagnostic accuracy of CTseg segmentation software for dementia in Pacific peoples living in Aotearoa New Zealand

 

Why this matters
• Pacific peoples in AotearoaNew Zealand experience a disproportionate dementia burden.1
• They tend to develop dementia at a younger age and are often diagnosed later.2
• Routine CT scans are accessible; automated volumetry could make case-finding scalable.
• CTsegis an automated tool that extracts brain volumes from routine non-contrast CT scans.3
• CTseg has previously shown moderate accuracy for dementia case-finding in NZ Europeans.4

Objective
Evaluate whether CTseg-derived totalbrain volume and hippocampal volume distinguish dementia from non-dementia in Pacific patients attending an NZ memory service.

Study design
1
Population
204 Pacific patients aged 65yr+; consecutively selected from memory-service sample in South Auckland, 2013-2022.
2
Dementiaclassification
Based on DSM-5 criteria, subtype and severity determined by multidisciplinary clinical team consensus.
3
Imaging
Non-contrast routinely collected CT brain scan within six months of diagnosis
4

Analysis
ANCOVA and logistic regression adjusted for age and total intracranial volume; ROC/AUC and Youden thresholds.
CTseg image-processing workflow
CTseg estimates TIV/TBV and tissue maps; AAL hippocampal masks are applied after normalization.

Abbreviations
TBV = total brain volume
HV = hippocampal volume
TIV = total intracranial volume
AD = Alzheimer's disease dementia; VD = vascular dementia
MCI = mild cognitive impairment

Cohort and diagnostic mix
204
Pacific patients with CT scans
127
Dementia
77
No dementia
72%
High deprivation
Pacificethnicities
Samoan
115
Cook Islands Māori
40
Tongan
37
Other Pacific
12
Dementia subtypes
AD dementia
56
Vasculardementia
56
Mixed dementia
15

Main finding
All CTseg-derived volumes were significantly lower in dementia after adjustment for age and TIV
Largest difference: hippocampal volume
ROC curves and diagnostic performance

Threshold choice matters
Higher probability thresholds increased specificity and PPV, but reduced sensitivity and NPV. CTseg is therefore better positioned as a triage/case-finding support tool, not a standalone diagnostic test.
What the findings mean
Moderate accuracy, real-world sample
CTseg performed comparably to the previous NZ European memory-service sample. Confidence intervals overlapped, suggesting no loss of performance in this Pacific sample.
Strengths
• First CT-based volumetry study focused on a Pacific dementia cohort.
• Consecutive, routine clinical sample.
• Diagnosis made by multidisciplinary consensus.
• Uses routinely acquired CT scans –potential for opportunistic identification of brain changes in CT scans conducted for other reasons
Limitations
• CT scans were viewed used for clinical dementia assessment so incorporation bias may have occurred.
• Small non-dementia group and class imbalance could affect specificity.
• Model performance indicates CT volumes is not ready for standalone clinical use
Conclusions
CTseg-derived brain volumes can identify dementia with moderate accuracy in Pacific peoples and may support earlier, more equitable dementia case-finding when integrated into multimodal clinical workflows.5
Next step: combine CTseg outputs with routinely collected electronic health data and validate against a gold standard dementia assessment.

Contact
Mukish Yelancheziany.mukish.m.Yelanchezian@aucklanduni.ac.nz
School of Medicine, University of Auckland, New Zealand
Scan for DOI
.