Spaces:
Running
Running
File size: 18,754 Bytes
79258bc d5410bd 79258bc 4027499 d5410bd 79258bc d5410bd 79258bc 4027499 d5410bd 79258bc d5410bd 79258bc a4b4765 79258bc d5410bd 79258bc d5410bd 79258bc d5410bd 79258bc 4027499 79258bc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 | """
SGP Parcellation Module
=======================
Maps TRIBE v2 fsaverage5 cortical output (~20,484 vertices)
to the 9 SGP nodes using the Schaefer-200 atlas ROI assignments.
Node definitions are grounded in the dual-stream language model
and Human Connectome Project tractography literature.
Nodes:
G1 Broca - Phonological production, syntactic processing
G2 Wernicke - Auditory comprehension, lexical-semantic decoding
G3 TPJ - Stream convergence, sensorimotor interface
G4 PFC - Executive control, coherence/veto
G5 DMN - Generativity, self-referential processing
G6 Limbic - Emotional weighting, memory consolidation
G7 Sensory - Primary perceptual input encoding
G8 ATL - Cross-modal semantic integration hub
G9 Premotor - Action planning, motor speech preparation
White Matter Tracts (edge weights derived from co-activation):
AF Arcuate Fasciculus G2 <-> G1
SLF Superior Long. Fasc. G3 <-> G4
IFOF Inf. Fronto-Occip. Fasc. G8 <-> G4
ILF Inf. Long. Fasc. G7 <-> G2
UF Uncinate Fasciculus G8 <-> G6
CG Cingulum G6 <-> G4 <-> G5
CC Corpus Callosum bilateral integration
MdLF Mid. Long. Fasc. G2 <-> G7
"""
import numpy as np
from typing import Dict, Tuple
import os
import urllib.request
# βββ Schaefer-200 ROI to SGP Node Mapping ββββββββββββββββββββββββββββββββββββ
# The Schaefer-200 atlas divides the cortex into 200 parcels (100/hemisphere).
# Each parcel belongs to one of 7 Yeo networks. We map these to our 9 SGP nodes.
#
# Yeo network assignments:
# Vis = Visual
# SomMot = Somatomotor
# DorsAttn = Dorsal Attention
# SalVentAttn = Salience/Ventral Attention
# Limbic = Limbic
# Cont = Control (Frontoparietal)
# Default = Default Mode
#
# Additional anatomical assignments for language regions not captured by Yeo:
# IFG pars opercularis/triangularis -> G1 Broca
# posterior STG/STS -> G2 Wernicke
# TPJ (supramarginal + angular gyrus) -> G3
# anterior temporal lobe -> G8 ATL
# Schaefer-200 parcel indices (1-indexed, as in atlas) mapped to SGP nodes.
# Left hemisphere: parcels 1-100, Right hemisphere: 101-200
# These assignments are based on Yeo network membership + anatomical location.
SGP_NODE_DEFINITIONS = {
"G1_broca": {
"description": "Broca's Area - phonological production, syntactic processing",
"stream": "dorsal",
"hemisphere": "left_dominant",
# IFG pars opercularis (BA44) and triangularis (BA45)
# Schaefer-200 Cont network, inferior frontal
"yeo_networks": ["Cont"],
"anatomical_keywords": ["FrOperIns", "Broca", "Tri", "Oper"],
"mni_center": [-51, 12, 18], # approximate MNI center
},
"G2_wernicke": {
"description": "Wernicke's Area - auditory comprehension, lexical-semantic",
"stream": "ventral",
"hemisphere": "left_dominant",
# Posterior superior temporal gyrus (BA22), planum temporale
"yeo_networks": ["SalVentAttn"],
"anatomical_keywords": ["ParOper", "Wernicke"],
"mni_center": [-54, -36, 12],
},
"G3_tpj": {
"description": "Temporoparietal Junction - stream convergence, sensorimotor interface",
"stream": "convergence",
"hemisphere": "bilateral",
# Supramarginal gyrus + angular gyrus
"yeo_networks": ["DorsAttn", "SalVentAttn"],
"anatomical_keywords": ["DorsAttn_Post", "ParieTempOcc", "Angular"],
"mni_center": [-54, -42, 24],
},
"G4_pfc": {
"description": "Prefrontal Cortex - executive control, coherence/veto",
"stream": "dorsal",
"hemisphere": "bilateral",
# DLPFC, anterior cingulate, orbitofrontal
"yeo_networks": ["Cont"],
"anatomical_keywords": ["PFCl", "PFC", "Frontal", "ACC", "Cing"],
"mni_center": [-36, 48, 18],
},
"G5_dmn": {
"description": "Default Mode Network - generativity, self-referential processing",
"stream": "generative",
"hemisphere": "bilateral",
# Medial PFC, posterior cingulate, angular gyrus
"yeo_networks": ["Default"],
"anatomical_keywords": ["pCunPCC", "Default_Par", "PHC"],
"mni_center": [0, -52, 26],
},
"G6_limbic": {
"description": "Limbic System - emotional weighting, memory consolidation",
"stream": "modulatory",
"hemisphere": "bilateral",
# Amygdala, hippocampus, insula, parahippocampal
"yeo_networks": ["Limbic", "SalVentAttn"],
"anatomical_keywords": ["Limbic", "TempPole", "OFC", "Insula", "ParaHipp", "Hipp", "Amyg"],
"mni_center": [-24, -18, -18],
},
"G7_sensory": {
"description": "Primary Sensory Cortices - perceptual input encoding",
"stream": "input",
"hemisphere": "bilateral",
# V1/V2 (occipital), A1 (Heschl's gyrus), S1 (postcentral gyrus)
"yeo_networks": ["Vis", "SomMot"],
"anatomical_keywords": ["Vis", "SomMot", "Medial"],
"mni_center": [0, -84, 6],
},
"G8_atl": {
"description": "Anterior Temporal Lobe - cross-modal semantic integration",
"stream": "ventral",
"hemisphere": "bilateral",
# Temporal pole, anterior MTG, anterior ITG
"yeo_networks": ["Default", "Limbic"],
"anatomical_keywords": ["Default_Temp", "Cont_Temp"],
"mni_center": [-42, 6, -30],
},
"G9_premotor": {
"description": "Premotor/SMA - action planning, motor speech preparation",
"stream": "dorsal",
"hemisphere": "left_dominant",
# Premotor cortex, SMA, precentral gyrus inferior
"yeo_networks": ["SomMot", "Cont"],
"anatomical_keywords": ["FEF", "PrCv", "Precentral", "Motor"],
"mni_center": [-42, 0, 48],
},
}
# White matter tract definitions β edges in the Resonance Graph
SGP_TRACT_DEFINITIONS = {
"AF": {
"name": "Arcuate Fasciculus",
"connects": ("G2_wernicke", "G1_broca"),
"function": "Phonological loop, direct sound-to-production",
"stream": "dorsal",
},
"SLF": {
"name": "Superior Longitudinal Fasciculus",
"connects": ("G3_tpj", "G4_pfc"),
"function": "Speech planning, working memory relay",
"stream": "dorsal",
},
"IFOF": {
"name": "Inferior Fronto-Occipital Fasciculus",
"connects": ("G8_atl", "G4_pfc"),
"function": "Semantic/visual meaning stream (ventral)",
"stream": "ventral",
},
"ILF": {
"name": "Inferior Longitudinal Fasciculus",
"connects": ("G7_sensory", "G2_wernicke"),
"function": "Visual word form, perceptual-to-language",
"stream": "ventral",
},
"UF": {
"name": "Uncinate Fasciculus",
"connects": ("G8_atl", "G6_limbic"),
"function": "Emotional-semantic integration",
"stream": "modulatory",
},
"CG_exec": {
"name": "Cingulum (Executive)",
"connects": ("G6_limbic", "G4_pfc"),
"function": "Memory-executive balance",
"stream": "modulatory",
},
"CG_dmn": {
"name": "Cingulum (DMN)",
"connects": ("G4_pfc", "G5_dmn"),
"function": "Ego-creativity balance, DMN regulation",
"stream": "generative",
},
"CC": {
"name": "Corpus Callosum",
"connects": ("G3_tpj", "G3_tpj"), # bilateral β L<->R TPJ as proxy
"function": "Interhemispheric coherence",
"stream": "integration",
},
"MdLF": {
"name": "Middle Longitudinal Fasciculus",
"connects": ("G2_wernicke", "G7_sensory"),
"function": "Auditory processing continuity",
"stream": "ventral",
},
}
class SGPParcellator:
"""
Maps TRIBE v2 fsaverage5 vertex activations to SGP node scores.
Uses a vertex-to-parcel lookup table derived from the Schaefer-200 atlas
projected onto the fsaverage5 surface mesh.
"""
# fsaverage5 has 20,484 vertices (10,242 per hemisphere)
N_VERTICES = 20484
N_VERTICES_PER_HEMI = 10242
# Schaefer-200 atlas file URLs (fsaverage5 surface labels)
ATLAS_URLS = {
"lh": "https://raw.githubusercontent.com/ThomasYeoLab/CBIG/master/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/FreeSurfer5.3/fsaverage5/label/lh.Schaefer2018_200Parcels_7Networks_order.annot",
"rh": "https://raw.githubusercontent.com/ThomasYeoLab/CBIG/master/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/FreeSurfer5.3/fsaverage5/label/rh.Schaefer2018_200Parcels_7Networks_order.annot",
}
def __init__(self, cache_dir: str = "/tmp/sgp_atlas"):
self.cache_dir = cache_dir
os.makedirs(cache_dir, exist_ok=True)
self._vertex_to_node = None # lazy load
def _build_vertex_to_node_map(self) -> np.ndarray:
"""
Build a (20484,) array mapping each vertex index to a SGP node name.
Uses keyword matching against Schaefer-200 parcel labels.
Falls back to Yeo network assignment if no keyword match.
"""
try:
import nibabel as nib
vertex_map = np.full(self.N_VERTICES, "G5_dmn", dtype=object) # DMN as default
for hemi_idx, hemi in enumerate(["lh", "rh"]):
annot_path = os.path.join(self.cache_dir, f"{hemi}.schaefer200.annot")
if not os.path.exists(annot_path):
print(f"[SGP] Downloading Schaefer-200 {hemi} atlas...", flush=True)
try:
urllib.request.urlretrieve(self.ATLAS_URLS[hemi], annot_path)
except Exception as e:
print(f"[SGP] Atlas download failed: {e}. Using fallback.", flush=True)
return self._fallback_vertex_map()
labels, ctab, names = nib.freesurfer.read_annot(annot_path)
# labels: (10242,) int array, each value is parcel index
# names: list of parcel name bytes
vertex_offset = hemi_idx * self.N_VERTICES_PER_HEMI
for v_local, parcel_idx in enumerate(labels):
v_global = v_local + vertex_offset
if parcel_idx <= 0 or parcel_idx >= len(names):
vertex_map[v_global] = "G7_sensory" # medial wall -> sensory fallback
continue
parcel_name = names[parcel_idx].decode("utf-8", errors="ignore")
vertex_map[v_global] = self._parcel_name_to_node(parcel_name, hemi)
print(f"[SGP] Parcellation map built. Node counts:", flush=True)
for node in SGP_NODE_DEFINITIONS:
count = np.sum(vertex_map == node)
print(f" {node}: {count} vertices", flush=True)
return vertex_map
except ImportError:
print("[SGP] nibabel not available. Using fallback parcellation.", flush=True)
return self._fallback_vertex_map()
def _parcel_name_to_node(self, parcel_name: str, hemi: str) -> str:
"""
Map a Schaefer-200 parcel name to an SGP node.
Priority: anatomical keyword match > Yeo network match > default (DMN).
"""
name_upper = parcel_name.upper()
# Check anatomical keywords first (highest specificity)
for node_id, node_def in SGP_NODE_DEFINITIONS.items():
for kw in node_def["anatomical_keywords"]:
if kw.upper() in name_upper:
return node_id
# Fall back to Yeo network
yeo_to_node = {
"VIS": "G7_sensory",
"SOMMOT": "G9_premotor",
"DORSATTN": "G3_tpj",
"SALVENTATT": "G2_wernicke",
"LIMBIC": "G6_limbic",
"CONT": "G4_pfc",
"DEFAULT": "G5_dmn",
}
for yeo_key, node_id in yeo_to_node.items():
if yeo_key in name_upper:
return node_id
return "G5_dmn" # final fallback
def _fallback_vertex_map(self) -> np.ndarray:
"""
Anatomically-motivated fallback when atlas files unavailable.
Divides fsaverage5 vertices by approximate cortical location.
This is a rough approximation β atlas-based mapping is strongly preferred.
"""
print("[SGP] Using fallback vertex map (approximate)", flush=True)
vertex_map = np.empty(self.N_VERTICES, dtype=object)
# Left hemisphere (vertices 0-10241): language dominant
lh_assignments = [
(0, 600, "G7_sensory"), # occipital/visual
(600, 1200, "G2_wernicke"), # posterior temporal
(1200, 1800, "G3_tpj"), # temporoparietal
(1800, 2400, "G6_limbic"), # medial temporal/limbic
(2400, 3000, "G8_atl"), # anterior temporal
(3000, 3600, "G1_broca"), # inferior frontal
(3600, 4200, "G4_pfc"), # prefrontal
(4200, 4800, "G9_premotor"), # premotor
(4800, 5400, "G5_dmn"), # medial/posterior (DMN)
(5400, 10242, "G5_dmn"), # remaining -> DMN
]
for start, end, node in lh_assignments:
vertex_map[start:end] = node
# Right hemisphere (vertices 10242-20483): less language dominant
rh_offset = self.N_VERTICES_PER_HEMI
rh_assignments = [
(0, 600, "G7_sensory"),
(600, 1200, "G7_sensory"), # bilateral visual
(1200, 1800, "G3_tpj"), # bilateral TPJ
(1800, 2400, "G6_limbic"),
(2400, 3000, "G8_atl"),
(3000, 3600, "G4_pfc"), # bilateral PFC
(3600, 4200, "G4_pfc"),
(4200, 4800, "G9_premotor"),
(4800, 10242, "G5_dmn"),
]
for start, end, node in rh_assignments:
vertex_map[rh_offset + start:rh_offset + end] = node
return vertex_map
def get_vertex_map(self) -> np.ndarray:
"""Lazy-load the vertex-to-node map."""
if self._vertex_to_node is None:
self._vertex_to_node = self._build_vertex_to_node_map()
return self._vertex_to_node
def parcellate(self, pred_array: np.ndarray) -> Dict:
"""
Convert TRIBE v2 prediction array to SGP node activation scores.
Args:
pred_array: (n_timesteps, n_vertices) float array from TRIBE v2
Returns:
dict with keys:
sgp_nodes: {node_id: float} activation score 0-1 per node
streams: {stream_name: float} mean activation per stream
edge_weights: {tract_id: float} co-activation strength per tract
dominant_hemisphere: "left" | "right" | "bilateral"
raw_stats: additional diagnostic info
"""
if pred_array.ndim == 1:
pred_array = pred_array.reshape(1, -1)
n_timesteps, n_vertices = pred_array.shape
vertex_map = self.get_vertex_map()
# Compute mean absolute activation per node
node_activations = {}
node_vertex_counts = {}
for node_id in SGP_NODE_DEFINITIONS:
mask = vertex_map[:n_vertices] == node_id
if mask.sum() == 0:
node_activations[node_id] = 0.0
node_vertex_counts[node_id] = 0
continue
node_act = float(np.abs(pred_array[:, mask]).mean())
node_activations[node_id] = node_act
node_vertex_counts[node_id] = int(mask.sum())
# Normalize to 0-1 range across nodes
max_act = max(node_activations.values()) if node_activations else 1.0
if max_act == 0:
max_act = 1.0
sgp_nodes = {
node_id: round(float(val / max_act), 4)
for node_id, val in node_activations.items()
}
# Stream aggregations
stream_map = {
"dorsal": ["G1_broca", "G3_tpj", "G4_pfc", "G9_premotor"],
"ventral": ["G2_wernicke", "G7_sensory", "G8_atl"],
"generative": ["G5_dmn"],
"modulatory": ["G6_limbic"],
"convergence": ["G3_tpj"],
}
streams = {
stream: round(float(np.mean([sgp_nodes[n] for n in nodes])), 4)
for stream, nodes in stream_map.items()
}
# Edge weights: co-activation strength between connected nodes
edge_weights = {}
for tract_id, tract_def in SGP_TRACT_DEFINITIONS.items():
n1, n2 = tract_def["connects"]
if n1 == n2:
# Bilateral tract (CC) β use hemispheric correlation
mid = n_vertices // 2
lh_act = float(np.abs(pred_array[:, :mid]).mean())
rh_act = float(np.abs(pred_array[:, mid:]).mean())
co_act = 1.0 - abs(lh_act - rh_act) / max(lh_act + rh_act, 1e-8)
else:
a1 = node_activations.get(n1, 0.0)
a2 = node_activations.get(n2, 0.0)
# Geometric mean as co-activation measure
co_act = float(np.sqrt(a1 * a2)) / max_act
edge_weights[tract_id] = round(co_act, 4)
# Hemispheric dominance
mid = min(n_vertices // 2, self.N_VERTICES_PER_HEMI)
lh_mean = float(np.abs(pred_array[:, :mid]).mean())
rh_mean = float(np.abs(pred_array[:, mid:n_vertices]).mean())
dom_ratio = (lh_mean - rh_mean) / max(lh_mean + rh_mean, 1e-8)
if dom_ratio > 0.05:
dominant_hemisphere = "left"
elif dom_ratio < -0.05:
dominant_hemisphere = "right"
else:
dominant_hemisphere = "bilateral"
return {
"sgp_nodes": sgp_nodes,
"streams": streams,
"edge_weights": edge_weights,
"dominant_hemisphere": dominant_hemisphere,
"raw_stats": {
"n_timesteps": n_timesteps,
"n_vertices": n_vertices,
"overall_mean_activation": round(float(np.abs(pred_array).mean()), 6),
"node_vertex_counts": node_vertex_counts,
"lh_mean": round(lh_mean, 6),
"rh_mean": round(rh_mean, 6),
}
}
# Module-level singleton
_parcellator = None
def get_parcellator(cache_dir: str = "/tmp/sgp_atlas") -> SGPParcellator:
global _parcellator
if _parcellator is None:
_parcellator = SGPParcellator(cache_dir=cache_dir)
return _parcellator
|