Behind the Veil
two
2
[Montana, 2024]

Reconstructing Perception from Brain Data

Consciousness Research & Statistical Modeling — Senior Thesis Montana State University

Overview
one
1
My thesis examined whether the phenomenology of immersive altered states could be structurally decoded from neural activity using Bayesian Estimation Image Reconstruction (BEIR). DMT was selected because it reliably produces immersive altered states with recurring phenomenological structure across participants.

Rather than relying solely on subjective reports, the study modeled whether simulated EEG and fMRI data could be translated into reconstructed imagery using a pre-trained deep neural network decoder. The central question was whether image reconstruction techniques could provide a quantifiable method for analyzing altered states of consciousness. 
one
Presentation poster for thesis
two
2
two
Model & Design
three
3
The study followed a within-subjects design (N=20, simulated dataset) modeled from established neuroimaging protocols. Neural data (EEG and fMRI) were processed through the BEIR framework, which integrates pre-trained Deep Neural Networks (VGG19, VQGAN, CLIP) to decode semantic visual content from brain activity.
‍
Statistical validation included: Repeated-measures ANOVA to compare global brain connectivity under DMT and placebo conditions. Linear regression to assess relationships between neural activation and image reconstruction quality. Finally, a Pearson correlation between reconstructed image coherence and MEQ-30 subjective intensity scores.
three
Key Findings
four
4
  • Global brain connectivity significantly differed between modeled DMT and placebo conditions (F(1,19) = 15.62, p < .001).
  • Reconstruction coherence showed a positive linear relationship with neural activation (p < .05).
  • MEQ-30 intensity scores strongly correlated with reconstructed image quality (r(18) = .837, p < .01).
  • Bayesian estimation improved semantic reconstruction under partial neural input, suggesting viability for modeling complex mental imagery.
four
Limitations & Implications
The dataset was simulated and designed to evaluate methodological viability rather than clinical outcomes. Reconstruction fidelity remains constrained by current fMRI spatial resolution and decoder architecture limits. 

Even with these constraints, this project demonstrates a method for bridging neural data and subjective phenomenology. Beyond psychedelic states, this framework has implications for dream research, memory reconstruction, and computational modeling of altered perception.

At a broader level, the thesis reflects an interdisciplinary approach to apply machine learning frameworks to questions traditionally confined to philosophy of mind and phenomenology. It examines how far emerging computational tools can extend into domains that have traditionally been confined to philosophy and self-report.
five
five
5
On Record
The Full Report [Dive Deeper]
Presentation Poster [View Full Poster]
six
six
6
← Previous Project [Investigative Article]
⌂ Projects [Full Library]
Next Project → [Digital Marketing]