Article
3D computational neuroimaging via
slab photography and deep learning
https
://
doi.org
/
10.1038
/
Received
:
21 May 2024
Accepted
:
Published online
:
Check for updates
Sean I. Young
1,2 ✉
, Juan Eugenio Iglesias
1,2
& Bruce Fischl
1
Traditional neuroimaging relies on MRI and histological sectioning as a reference for
postmortem reconstruction of brain structures, limiting the utility and scalability of
studies in which either modality is unavailable. Here, we present a 3D computational
neuroimaging modality based entirely on photography of brain slabs. Our approach
is based on self-supervised deep learning for 3D volume reconstruction, enabling an
accurate spatial map between individual brain slabs and underlying brain without the
need for an additional imaging modality, such as MRI. Te s t e d o n p o s t m o r t e m b r a i n
specimens, this approach leads to a higher level of resolution and accuracy compared
to MRI-based methods while maintaining anatomical integrity. This computational
imaging modality offers a cost-effective and non-invasive alternative for postmortem
brain imaging, with potential applications in neuropathology and neurology, as well
as forensic investigations. By eliminating dependence on expensive imaging systems
such as MRI scanners, this neuroimaging modality democratizes neuroimaging and
enables new avenues of large-scale studies of neuroimaging–pathology correlations
through retrospective reconstruction of historical and current slab photographs for
neurological conditions of interest, such as Alzheimer’s disease.
A list of affiliations appears at the end of the paper.
Nature | www.nature.com | 28 November 2025 | 1
Neuroimaging plays a key role in correlating the brain’s structures and
its pathology, with magnetic resonance imaging (MRI) and histological
sectioning serving as the “gold-standard” techniques for postmortem
brain reconstruction. Although these modalities can provide valuable
insights into neurodegenerative diseases, traumatic brain injury, and
structural abnormalities, they come with significant limitations. MRI,
though widely used, offers limited histological resolution and requires
expensive infrastructure, rendering it inaccessible for many research
environments. Logistical issues can further complicate both cadeveric
and ex vivo MRI of the brain, as tissue degradation due to autolysis and
bacteria introduces artifacts in acquired MR images. These constraints
significantly restrict the applicability and accessibility of postmortem
neuroimaging, particularly if only slab photographs are available, such
as in the case of historical brain banks. Several recent works attempt to
reconstruct brain models from a combination of slab photographs and
MRI (or brain surface scans) using numerical methods
1,2
. However, slab
photography as a stand-alone neuroimaging modality on par with MRI
has remained elusive due to the computational challenges of piecing
together different brain slabs, each of which is subjected to geometric
distortions from tissue deformation as well as camera distortion.
Here, we introduce a computational neuroimaging modality that is
capable of imaging high-fidelity 3D brain volumes from conventional
photographs of postmortem brain slabs, eliminating the need for MRI
and other imaging modalities. Our method leverages fully supervised
deep learning to learn spatial correspondences between 2D brain slab
images and the underlying 3D brain, enabling an accurate volumetric
reconstruction without additional imaging modalities. By leveraging
deep learning’s ability to infer complex spatial structures from limited
textural cues, this approach achieves reconstruction accuracies which
surpass MRI-based methods while preserving anatomical integrity. We
test our approach on postmortem brain specimens, demonstrating its
ability to reconstruct 3D brains with high spatial fidelity. Our findings
highlight the potential of deep-learning-driven photographic imaging
as a cost-effective, standalone computational imaging modality for 3D
postmortem neuroimaging. This paradigm shift will enable large-scale
studies utilizing existing slab photographs from brain banks to unlock
new avenues for neuropathology and neurodegeneration studies, and
forensic investigation. By removing reliance on conventional imaging
systems, our method democratizes high-resolution neuroimaging for
the studies of neurological conditions and diseases, and facilitates new
insights into their structural underpinnings.
This approach not only democratizes high-resolution neuroimaging
for the fields of neuropathology and forensic science, but also paves
the way for a range of future applications. For example, it can enable
large-scale retrospective studies of neurodegenerative diseases by
transforming archived slab photographs into three-dimensional data,
thus unlocking the potential of existing brain banks. In the realm of
forensic investigations, this technique offers the ability to create
accurate 3D reconstructions of injury patterns from autopsy photos,
providing courts and investigators with a new level of spatial detail.
Moreover, it allows researchers to revisit historical neurological
archives with modern analytical techniques, fostering new insights
into the structural underpinnings of neurological conditions. In
summary, this innovation extends the reach of 3D neuroimaging into
new domains and facilitates a broader understanding of the brain
through purely photographic data.
Conventional slab photography
2D slab photographs of the brain are routinely captured at brain banks
for postmortem neuroimaging, providing a detailed visual record of a
brain’s anatomy for neuropathological assessments as well as forensic
investigation. Unlike in vivo neuroimaging modalities such as MRI and
CT, slab photography can capture high-resolution surface detail of the
dissected brain tissue, enabling direct visualization of morphological
changes due to neurodegeneration, traumatic injuries, and pathology
in the vasculature. However, slab photography is a destructive form of
imaging—brain dissection is an irreversible process as slabs undergo a
