MRI-derived Face Age vs Brain Age from the Same T1 Scan: a Proof-of-Concept Pipeline

Ekaterina Kondrateva, Ramil Khafizov, Gleb Bobrovskikh

Keywords: brain age, face age, MRI, biological age, age gap, reproducibility

TL;DR: We compare MRI-derived face-age and brain-age estimates from the same T1 scan to test whether these two aging signals agree or carry independent biological information.

Registration Requirement: Yes

Abstract:

A non-defaced T1-weighted MRI contains two distinct aging signals within a single volume: the brain parenchyma and the full facial anatomy. We present an open pipeline that extracts both signals in parallel from the same scan by combining a brain-age model (SynthBA) with two face-age approaches: multi-view surface renders scored by FaceAge and morphometric features derived from BioFace3D-20 landmarks.

We evaluated the pipeline on the SIMON single-subject dataset, which includes 99 scans acquired across 36 scanners between ages 29 and 46 years. The results reveal a critical failure mode. Despite strong scan-to-scan consistency, SynthBA produced an almost constant predicted age of 27.4 ± 1.26 years across the entire series, yielding a mean absolute error of 16.2 years. This shows that high scanner reproducibility can coexist with severe out-of-distribution bias and poor validity.

In contrast, the facial measures retained a detectable relationship with chronological age. Face morphometrics remained positively biased by 5.1 years on average but showed a clear age-related trend, while FaceAge predictions from MRI-derived renders showed a larger positive bias of 8.5 years, consistent with the domain gap between MRI-based surface renderings and natural photographic images.

The main finding is methodological rather than purely predictive: reproducibility should not be interpreted as evidence of validity. Our results suggest that facial geometry extracted from the same MRI scan may preserve longitudinal aging information that current brain-age models fail to capture.

Reproducibility: https://github.com/kondratevakate/faceage-to-brainage

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 99