Treatment / Current direction
Link fMRI patterns to treatment response.
Can brain signals add to clinical evidence?
Image patternsClinical outcomesExternal validation
AI × functional brain imaging
We study fMRI images with AI to understand how brain networks work together—and what changes in mental illness.
From scan to insight From scan to insight
Published motor-contrast mapNeuroVault / 10426
NeuroAI / Functional image research
Read the image.
Understand the system.
AI image recognition.
Brain networks in context.
3D slice stack
axialz = 37 mm
Positive contrastNegative contrastSpatial review · not a live acquisition
01 / Research approach
A volume, slice by slice.
Read functional images in their spatial context. Quality and alignment come first.
3DSpatial image3 mmVoxel spacing
This published motor contrast shows the image layer. Our research extends to 4D fMRI: space plus time.
Image review / spatial quality control
Published fMRI contrast + reference anatomy. The animation shows image review, not AI inference or a diagnostic result.
View the original imageOur direction
AI is the method.
Psychiatric evidence is the goal.
01 / Research
One aim.
Three routes.
02 / Open evidence
Small circuits.
Bigger questions.
Research lens / MaleCNS
A nervous system,
cell by cell.
Google Research and collaborators mapped the male fruit fly’s brain and ventral nerve cord: more than 166,000 neurons and 125 million synaptic connections. AI traces cells from electron-microscope images; experts review the reconstructed anatomy.
We reanalyse the released annotations and neuron geometry below, then set out how network-based thinking can inform our psychiatric research.
Google Research · 03 Sep 2026Original study / Cell01Electron microscopyImage the tissue
02AI segmentationTrace neuronal shapes
03Expert proofreadingReview the reconstruction
04ConnectomeAnalyse the wiring
The Google article describes flood-filling convolutional networks that iteratively extend a cell segmentation through 3D image space. Human proofreading remains essential. Our analysis begins with the released tables and skeletons, after reconstruction.
Public-data analysis / MaleCNS v1.0
165,122 neurons in this analysis
Filtered to status = Traced
Recomputed 05 Sep 2026
Where the neurons belong
Released superclass annotations · complete filtered cohort
- Optic-lobe intrinsic89,39054.1%
- Central-brain intrinsic32,16019.5%
- Ventral-cord intrinsic13,1518.0%
- Visual projection9,2015.6%
- Ventral-cord sensory6,3653.9%
- Central-brain sensory4,8682.9%
- Other annotated classes9,4715.7%
- Unassigned5160.3%
Denominator: 165,122 neurons with status = Traced. Bars share a fixed 0–100% scale; counts are neurons, not synapses.
Data: FlyEM (HHMI Janelia), University of Cambridge, MRC Laboratory of Molecular Biology and Google Research. CC BY 4.0. Aggregation and centreline-display adaptation: NeuroAI. Dataset & release files.
What the original study adds
Local differences.
Network-wide questions.
The research team reports 262 sex-specific and 114 sexually dimorphic cell types. Many differences sit in higher-order centres, illustrating why circuitry matters beyond a simple neuron count.
For NeuroAI, the methodological lesson is to examine network organisation alongside individual features—not to equate fly circuits with human psychiatric diagnoses.
Janelia / published findingsNeuroAI / proposed study
Functional images.
Clinical questions.
Can AI recognise patterns in 4D fMRI that help explain treatment response and symptom change?
This is the human fMRI study design we want to test. The results above describe a fly dataset; they do not demonstrate a psychiatric biomarker or treatment benefit.
Discuss the research questionStructural connectomicsNeuron → neuron
EM reconstruction measures anatomical synaptic wiring.
Functional connectivityRegion ↔ region
fMRI estimates statistical relationships between regional BOLD signals.
Clinical first
Begin with a psychiatric question, not an algorithm.
Longitudinal
Model change rather than a single snapshot.
Transportable
Test evidence beyond one site or dataset.
Accountable
Keep assumptions visible and challengeable.
Research context
Technical ambition,
clinically examined.
Independent programme · Not an official MIT website03 / People
Six perspectives.
One programme.
MIT × Google · 2022–2025
Tomohisa Sasaki
Brain imaging AI research
Participated in a joint MIT–Google research project developing AI to quantify psychiatric states from brain-imaging data. Worked across image preprocessing and analysis, model implementation, performance and generalisation testing, and communication of findings.
- Programme
- MIT × Google joint research
- Period
- Jun 2022 – Feb 2025
- Domain
- AI for psychiatric imaging
04 / Collaborate
Bring the question.
Build the evidence.
Selected context