CALLIOPE · BCOM

This technical note documents the assignment of four annotation columns to each of the 84 parcels in the cortical_type_full_annotations.csv atlas file, which combines the Desikan--Killiany (DK) cortical parcellation with eight bilateral subcortical structures and bilateral cerebellum. The four columns are: (1)~Cortical Type (cytoarchitectonic class, pre-existing), (2)~Agent Model module ( {}, {}, {}) from the KT algorithmic agent framework of Ruffini et al.\ , (3)~Brodmann Area(s) from classical cytoarchitectonic mapping , (4)~Clinical Region — a set of 37 named anatomical shorthand labels used in the clinical neuroscience literature, and (5)~Yeo 7-Network from the Yeo et al.\ 2011 resting-state parcellation . (6)~MNI152 centroid coordinates (pos_x, pos_y, pos_z) in millimetres. For each annotation scheme we state the source, the assignment rules, and all cases where the assignment is ambiguous or represents a majority-vote over a continuous boundary.

A meticulously cross-referenced lookup table that maps every region of the Desikan-Killiany brain atlas to four parallel annotation systems simultaneously, designed to support computational psychiatry research on depression.

The core problem this solves is a coordination failure. Neuroscientists, clinicians, and computational modelers all carve the brain up differently. A neuroimager thinks in Yeo resting-state networks (DMN, FPN, etc.). A clinician thinks in anatomical shorthand (sgACC, dlPFC). A cytoarchitectonics person thinks in Brodmann areas. A theorist working from Ruffini et al.'s algorithmic agent framework thinks in terms of Modeling Engine, Objective Function, and Planning Engine — three abstract functional modules loosely analogous to a world-model, a reward evaluator, and an action planner. This document assigns all four vocabularies to each of 84 brain parcels (68 cortical + 16 subcortical), plus MNI152 centroid coordinates for spatial targeting, and packages everything into a single CSV file.

The agent-module layer is the most novel piece. The Modeling Engine (ME) covers regions that build internal representations of the world — posterior cortex, hippocampus, default mode network. The Objective Function (OF) covers regions that compute valence and reward — orbitofrontal cortex, subgenual ACC, amygdala, nucleus accumbens, thalamus. The Planning Engine (PE) covers regions that select and execute actions — dlPFC, premotor cortex, dorsal ACC, frontoparietal network. Of the 84 parcels, 38 are ME, 24 are OF, and 22 are PE. The document is careful about dual-role regions: the precuneus straddles ME and PE, the OFC straddles OF and PE, and these ambiguities are explicitly flagged rather than silently resolved.

The Yeo-to-agent-module crosswalk is particularly useful. DMN maps to ME, Limbic maps to OF, FPN and DAN map to PE — with high internal consistency (e.g., all 10 cortical Limbic parcels are OF, all 4 FPN parcels are PE). The one genuinely tricky case is the Ventral Attention Network, which splits across OF and PE depending on the specific region, reflecting the salience network's dual role in detecting affectively significant events and redirecting attention.

This is a reference document, not a research paper with novel empirical findings. Its value is infrastructural: it lets a researcher building a whole-brain model of major depressive disorder immediately know that, say, the subgenual ACC is OF/LIM/BA25/sgACC at coordinates (−6, 36, 8), and that this is the same region Mayberg targets with deep brain stimulation. Every ambiguous assignment is documented with its rationale, which is the right engineering discipline for a resource meant to be used and extended by others.