data/data_loader.py

# ==================== data/data_loader.py ====================
"""Data loading and preprocessing functionality"""

import pandas as pd
import numpy as np
from typing import List, Dict, Tuple, Optional
import os

class DataLoader:
    """Handles loading and parsing of similarity data"""
    
    def __init__(self):
        self.data: Optional[pd.DataFrame] = None
        self.ml_models: List[str] = []
        self.brain_measures: List[str] = []
        self.voxel_measures: List[str] = []
        # New categorization based on INPUT SOURCE + METHOD TYPE
        self.model_categories: Dict[str, List[Tuple[str, int]]] = {
            'vision': [],                  # Vision models (images)
            'captions_neural': [],         # Neural language models on captions
            'captions_statistical': [],    # Statistical text analysis on captions
            'tags_statistical': []         # Statistical text analysis on tags
        }
        # Encoder lookups
        self.roi_encoder: Optional[Dict] = None
        self.voxel_encoder: Optional[Dict] = None
        self.voxel_region_labels: Optional[List[str]] = None
    
    def get_model_type(self, model_name: str) -> str:
        """Categorize model types based on INPUT SOURCE + METHOD TYPE

        Categories:
        - vision: Models using image data (BOLD5000_timm_*, clip_*)
        - captions_neural: Neural language models on captions (bert, deberta, simcse, roberta)
        - captions_statistical: Statistical text analysis on captions (bm25, rouge, tf-idf, co-occurrence + _captions)
        - tags_statistical: Statistical text analysis on tags (bm25, rouge, tf-idf, co-occurrence + _tags)
        """
        # Strip _standardized suffix for categorization
        base_name = model_name.replace('_standardized', '')

        # Vision models - use raw images
        if "timm_" in base_name or "clip_" in base_name:
            return "vision"

        # Statistical text models on TAGS
        elif any(x in base_name for x in ["bm25", "rouge", "tf-idf", "co-occurrence"]) and "_tags" in base_name:
            return "tags_statistical"

        # Statistical text models on CAPTIONS
        elif any(x in base_name for x in ["bm25", "rouge", "tf-idf", "co-occurrence"]) and "_captions" in base_name:
            return "captions_statistical"

        # co-occurrence-rep_tags is also tags
        elif "co-occurrence-rep_tags" in base_name:
            return "tags_statistical"

        # Neural language models (on captions - assuming all use captions unless specified)
        elif any(x in base_name for x in ["bert", "deberta", "simcse", "roberta"]):
            return "captions_neural"

        else:
            # Default to vision if unclear
            return "vision"
    
    def load_csv(self, csv_path: str, roi_encoder_path: str = None, voxel_encoder_path: str = None) -> bool:
        """Load similarity data from CSV/TSV file and optional encoder files
        
        Args:
            csv_path: Path to main CSV/TSV file
            roi_encoder_path: Optional path to ROI encoder file
            voxel_encoder_path: Optional path to voxel encoder file
        """
        try:
            # Try to detect separator (tab or comma)
            print(f"Loading data from: {csv_path}")
            
            # Check file extension to guess separator
            if csv_path.endswith('.tsv'):
                separator = '\t'
                print("Detected TSV format (tab-separated)")
            elif csv_path.endswith('.csv'):
                # Try to auto-detect
                with open(csv_path, 'r') as f:
                    first_line = f.readline()
                    if '\t' in first_line:
                        separator = '\t'
                        print("Detected tab-separated format")
                    else:
                        separator = ','
                        print("Detected comma-separated format")
            else:
                # Default to comma
                separator = ','
                print("Using comma separator (default)")
            
            # Load the data
            self.data = pd.read_csv(csv_path, sep=separator)
            print(f"[OK] Loaded: {len(self.data)} rows, {len(self.data.columns)} columns")
            
            # Load encoder files if paths provided
            if roi_encoder_path:
                print(f"\n[ROI] Loading ROI encoder: {roi_encoder_path}")
                self.load_roi_encoder(roi_encoder_path)

            if voxel_encoder_path:
                print(f"[VOXEL] Loading voxel encoder: {voxel_encoder_path}")
                self.load_voxel_encoder(voxel_encoder_path)
            
            # Extract all components
            self._extract_ml_models()
            self._categorize_models()
            self._extract_brain_measures()
            self._compute_hierarchy_averages()
            self._print_summary()
            return True

        except Exception as e:
            print(f"[ERROR] Error loading data: {e}")
            import traceback
            traceback.print_exc()
            return False
    
    def load_roi_encoder(self, roi_path: str) -> bool:
        """Load ROI encoder lookup file
        
        Expected format:
        - Column 'image_filename' with image identifiers
        - Column 'avg_roi_across_subjects' with ROI arrays as strings
        """
        try:
            # Load CSV (your file is comma-separated)
            roi_df = pd.read_csv(roi_path)
            
            print(f"  Loaded {len(roi_df)} images from ROI encoder")
            print(f"  Columns: {list(roi_df.columns[:3])}... (total: {len(roi_df.columns)})")
            
            # Verify required columns exist
            if 'image_filename' not in roi_df.columns:
                raise ValueError(f"Column 'image_filename' not found. Available columns: {list(roi_df.columns)}")
            
            if 'avg_roi_across_subjects' not in roi_df.columns:
                raise ValueError(f"Column 'avg_roi_across_subjects' not found. Available columns: {list(roi_df.columns)}")
            
            # Create lookup: image_filename -> avg_roi_across_subjects
            self.roi_encoder = {}
            success_count = 0
            for _, row in roi_df.iterrows():
                img_name = row['image_filename']
                # Parse the averaged ROI string
                roi_str = row['avg_roi_across_subjects']
                roi_values = self._parse_roi_string(roi_str)
                if roi_values is not None:
                    self.roi_encoder[img_name] = roi_values
                    success_count += 1

            print(f"  [OK] Created ROI lookup for {success_count} images")
            return True

        except Exception as e:
            print(f"  [WARNING] Error loading ROI encoder: {e}")
            import traceback
            traceback.print_exc()
            self.roi_encoder = None
            return False
    
    def load_voxel_encoder(self, voxel_path: str) -> bool:
        """Load voxel encoder lookup file
        
        Expected format:
        - Column 'image_filename' with image identifiers
        - Columns for each ROI region (EBA, FBA2, FFA1, etc.)
        """
        try:
            # Load CSV (your file is comma-separated)
            voxel_df = pd.read_csv(voxel_path)
            
            print(f"  Loaded {len(voxel_df)} images from voxel encoder")
            print(f"  Columns: {list(voxel_df.columns[:5])}... (total: {len(voxel_df.columns)})")
            
            # Verify image_filename column exists
            if 'image_filename' not in voxel_df.columns:
                raise ValueError(f"Column 'image_filename' not found. Available columns: {list(voxel_df.columns)}")
            
            # Extract region labels (all columns except image_filename)
            self.voxel_region_labels = [col for col in voxel_df.columns if col != 'image_filename']
            print(f"  [OK] Found {len(self.voxel_region_labels)} voxel regions")
            print(f"    Regions: {self.voxel_region_labels}")

            # Create lookup: image_filename -> voxel values array
            self.voxel_encoder = {}
            for _, row in voxel_df.iterrows():
                img_name = row['image_filename']
                voxel_values = np.array([float(row[col]) for col in self.voxel_region_labels])
                self.voxel_encoder[img_name] = voxel_values

            print(f"  [OK] Created voxel lookup for {len(self.voxel_encoder)} images")
            return True

        except Exception as e:
            print(f"  [WARNING] Error loading voxel encoder: {e}")
            import traceback
            traceback.print_exc()
            self.voxel_encoder = None
            self.voxel_region_labels = None
            return False
    
    @staticmethod
    def _parse_roi_string(roi_str) -> Optional[np.ndarray]:
        """Parse ROI string like '[-0.366, -0.379, ...]' into numpy array"""
        try:
            if pd.isna(roi_str) or roi_str == '':
                return None
            # Remove brackets and split
            roi_str = str(roi_str).strip('[]')
            values = [float(x.strip()) for x in roi_str.split(',') if x.strip()]
            return np.array(values)
        except Exception as e:
            return None
    
    def get_roi_values(self, image_filename: str) -> Optional[np.ndarray]:
        """Get ROI values for an image"""
        if self.roi_encoder is None:
            return None
        return self.roi_encoder.get(image_filename, None)
    
    def get_voxel_values(self, image_filename: str) -> Optional[np.ndarray]:
        """Get voxel-per-region values for an image"""
        if self.voxel_encoder is None:
            return None
        return self.voxel_encoder.get(image_filename, None)
    
    def _extract_ml_models(self):
        """Extract ML model columns - prefers standardized versions"""
        # First, find all potential ML model columns (both original and standardized)
        all_ml_candidates = [col for col in self.data.columns if
                        'BOLD5000_timm_' in col or
                        'clip_' in col or
                        'bert-' in col or
                        'deberta-' in col or
                        'sup-simcse' in col or
                        'roberta' in col or
                        'bm25' in col.lower() or
                        'tf-idf' in col or
                        'rouge' in col or
                        'co-occurrence' in col]

        # Build a mapping: base_name -> best_column
        # Prefer standardized versions over original
        model_map = {}

        for col in all_ml_candidates:
            # Get base name (remove _standardized suffix if present)
            if col.endswith('_standardized'):
                base_name = col[:-13]  # Remove '_standardized'
                # Prefer standardized version
                model_map[base_name] = col
            else:
                # Only use original if standardized version doesn't exist yet
                if col not in model_map:
                    model_map[col] = col

        # Get the final list of columns to use
        self.ml_models = list(model_map.values())

        # Count how many are standardized
        standardized_count = sum(1 for col in self.ml_models if col.endswith('_standardized'))

        print(f"Found {len(self.ml_models)} ML model columns")
        if standardized_count > 0:
            print(f"  Using {standardized_count} standardized columns (z-scored)")
            print(f"  Using {len(self.ml_models) - standardized_count} original columns")
        else:
            print(f"  All columns are original (not z-scored)")
    
    def _categorize_models(self):
        """Categorize models by INPUT SOURCE + METHOD TYPE"""
        self.model_categories = {
            'vision': [],
            'captions_neural': [],
            'captions_statistical': [],
            'tags_statistical': []
        }
        for i, model in enumerate(self.ml_models):
            category = self.get_model_type(model)
            self.model_categories[category].append((model, i))
    
    def _extract_brain_measures(self):
        """Extract brain response columns (NEW naming convention)"""
        self.brain_measures = []

        # NEW: ROI-based measures with new naming convention
        # Pattern: roi_{metric}_{roi_type}_avg_{measure_type}
        roi_patterns = [
            'roi_cosine_common_avg_sim',
            'roi_cosine_early_avg_sim',
            'roi_cosine_late_avg_sim',
            'roi_pearson_common_avg_sim',
            'roi_pearson_early_avg_sim',
            'roi_pearson_late_avg_sim',
            'roi_cosine_common_avg_roi',
            'roi_pearson_common_avg_roi',
            'roi_cosine_early_avg_roi',
            'roi_pearson_early_avg_roi',
            'roi_cosine_late_avg_roi',
            'roi_pearson_late_avg_roi',
        ]

        for measure in roi_patterns:
            if measure in self.data.columns:
                self.brain_measures.append(measure)

        # NEW: Individual ROI region measures (simple difference)
        # Pattern: roi_single_{region_name}
        roi_single_cols = [col for col in self.data.columns if col.startswith('roi_single_')]
        self.brain_measures.extend(roi_single_cols)

        # NEW: Individual ROI voxel-pattern measures (correlation across voxels)
        # Pattern: roi_voxel_{metric}_{region_name}
        roi_voxel_cols = [col for col in self.data.columns if col.startswith('roi_voxel_')]
        self.brain_measures.extend(roi_voxel_cols)

        # NEW: Voxel-level measures
        # Pattern: voxel_{metric}_{what}
        voxel_cols = [col for col in self.data.columns if
                      col.startswith('voxel_') and not col.startswith('voxel_to_roi_')]
        self.voxel_measures = voxel_cols
        self.brain_measures.extend(voxel_cols)

        # NEW: Voxel-to-ROI measures
        # Pattern: voxel_to_roi_{metric}_{roi_type}_avg_{measure_type}
        voxel_to_roi_cols = [col for col in self.data.columns if col.startswith('voxel_to_roi_')]
        self.brain_measures.extend(voxel_to_roi_cols)

        print(f"Found {len(self.brain_measures)} brain measure columns")

    def _compute_hierarchy_averages(self):
        """Compute hierarchy analysis averages for early visual and late semantic regions"""

        # Define ROI groups (same as hierarchy analysis)
        EARLY_ROIS = ['V1v', 'V1d', 'V2v', 'V2d', 'V3v', 'V3d', 'hV4']
        LATE_ROIS = ['mfswords', 'VWFA1', 'VWFA2', 'PPA', 'OPA', 'RSC',
                     'OWFA', 'FFA1', 'FFA2', 'OFA', 'EBA', 'FBA2']

        # Get columns for each group
        early_cols = [f'roi_voxel_pearson_{roi}' for roi in EARLY_ROIS
                      if f'roi_voxel_pearson_{roi}' in self.data.columns]
        late_cols = [f'roi_voxel_pearson_{roi}' for roi in LATE_ROIS
                     if f'roi_voxel_pearson_{roi}' in self.data.columns]

        if early_cols:
            # Compute average across early visual ROIs
            self.data['hierarchy_early_visual_avg'] = self.data[early_cols].mean(axis=1)
            self.brain_measures.append('hierarchy_early_visual_avg')
            print(f"[HIERARCHY] Created Early Visual Average ({len(early_cols)} ROIs)")

        if late_cols:
            # Compute average across late semantic ROIs
            self.data['hierarchy_late_semantic_avg'] = self.data[late_cols].mean(axis=1)
            self.brain_measures.append('hierarchy_late_semantic_avg')
            print(f"[HIERARCHY] Created Late Semantic Average ({len(late_cols)} ROIs)")

        # Compute average across ALL ROIs (early + late combined)
        all_roi_cols = early_cols + late_cols
        if all_roi_cols:
            self.data['hierarchy_all_rois_avg'] = self.data[all_roi_cols].mean(axis=1)
            self.brain_measures.append('hierarchy_all_rois_avg')
            print(f"[HIERARCHY] Created All ROIs Average ({len(all_roi_cols)} ROIs: {len(early_cols)} early + {len(late_cols)} late)")

    def _print_summary(self):
        """Print data loading summary"""
        print(f"\n{'='*60}")
        print(f"DATA LOADING SUMMARY")
        print(f"{'='*60}")
        print(f"Total image pairs: {len(self.data)}")
        print(f"\nML Models: {len(self.ml_models)} total (categorized by input source)")

        # Define display names
        category_labels = {
            'vision': 'Vision Models (Images)',
            'captions_neural': 'Neural Language Models (Captions)',
            'captions_statistical': 'Statistical Text Analysis (Captions)',
            'tags_statistical': 'Statistical Text Analysis (Tags)'
        }

        for category in ['vision', 'captions_neural', 'captions_statistical', 'tags_statistical']:
            if self.model_categories[category]:
                print(f"  {category_labels[category]}: {len(self.model_categories[category])}")

        print(f"\nBrain Measures: {len(self.brain_measures)} total")
        
        # Count by type
        roi_count = len([m for m in self.brain_measures if m.startswith('roi_')])
        voxel_count = len([m for m in self.brain_measures if m.startswith('voxel_') and not m.startswith('voxel_to_roi_')])
        voxel_to_roi_count = len([m for m in self.brain_measures if m.startswith('voxel_to_roi_')])
        
        print(f"  ROI measures: {roi_count}")
        print(f"  Voxel measures: {voxel_count}")
        print(f"  Voxel-to-ROI measures: {voxel_to_roi_count}")
        
        # Encoder status
        print(f"\nEncoder Files:")
        print(f"  ROI encoder: {'[OK] Loaded' if self.roi_encoder else '[X] Not found'}")
        if self.roi_encoder:
            print(f"    {len(self.roi_encoder)} images available")
        print(f"  Voxel encoder: {'[OK] Loaded' if self.voxel_encoder else '[X] Not found'}")
        if self.voxel_encoder:
            print(f"    {len(self.voxel_encoder)} images, {len(self.voxel_region_labels)} regions")
        
        print(f"{'='*60}\n")
    
    def get_ml_model_options(self) -> List[Tuple[str, any]]:
        """Get ML model options with averages and complete names"""
        options = []

        # Define category display names
        category_labels = {
            'vision': 'Vision Models',
            'captions_neural': 'Language Models',
            'captions_statistical': 'Statistical Text Models (Captions)',
            'tags_statistical': 'Statistical Text Models (Tags)'
        }

        # Add category averages section
        options.append(("CATEGORY AVERAGES", "header_averages"))

        category_order = ['vision', 'captions_neural', 'captions_statistical', 'tags_statistical']
        for category in category_order:
            if self.model_categories[category]:
                options.append((f"AVERAGE - {category_labels[category]}", f"avg_{category}"))

        # Add separator
        if any(self.model_categories.values()):
            options.append(("────────────────────────────────", "separator"))

        # ========== IMAGE-BASED MODELS ==========
        if self.model_categories['vision']:
            options.append(("─────────── IMAGE MODELS ───────────", "header_image_models"))

            for model_name, model_idx in self.model_categories['vision']:
                # Clean display name (remove _standardized for display)
                display_name = model_name.replace('_standardized', '')
                if model_name.endswith('_standardized'):
                    display_name += " [Z]"  # Indicate it's z-scored
                options.append((display_name, model_idx))

        # ========== CAPTION-BASED MODELS ==========
        if self.model_categories['captions_neural'] or self.model_categories['captions_statistical']:
            options.append(("─────────── CAPTION MODELS ───────────", "header_caption_models"))

            # Language models (deep learning)
            if self.model_categories['captions_neural']:
                options.append(("LANGUAGE MODELS", "header_captions_neural"))

                for model_name, model_idx in self.model_categories['captions_neural']:
                    # Clean display name
                    display_name = model_name.replace('_standardized', '')
                    if model_name.endswith('_standardized'):
                        display_name += " [Z]"
                    options.append((display_name, model_idx))

            # Statistical text models
            if self.model_categories['captions_statistical']:
                options.append(("STATISTICAL TEXT MODELS", "header_captions_statistical"))

                for model_name, model_idx in self.model_categories['captions_statistical']:
                    # Clean display name
                    display_name = model_name.replace('_standardized', '')
                    if model_name.endswith('_standardized'):
                        display_name += " [Z]"
                    options.append((display_name, model_idx))

        # ========== TAG-BASED MODELS ==========
        if self.model_categories['tags_statistical']:
            options.append(("─────────── TAG MODELS ───────────", "header_tag_models"))

            for model_name, model_idx in self.model_categories['tags_statistical']:
                # Clean display name
                display_name = model_name.replace('_standardized', '')
                if model_name.endswith('_standardized'):
                    display_name += " [Z]"
                options.append((display_name, model_idx))

        return options
    
    def get_brain_measure_options(self) -> List[Tuple[str, str]]:
        """Get brain measure options with clean names, organized by type"""
        options = []

        # ========== HIERARCHY ANALYSIS AVERAGES (NEW!) ==========
        if 'hierarchy_early_visual_avg' in self.data.columns or 'hierarchy_late_semantic_avg' in self.data.columns or 'hierarchy_all_rois_avg' in self.data.columns:
            options.append(("─────────── HIERARCHY ANALYSIS ───────────", "header_hierarchy"))
            options.append(("HIERARCHY AVERAGES (For Verification)", "header_hierarchy_avgs"))

            if 'hierarchy_early_visual_avg' in self.data.columns:
                options.append(("Early Visual Average (7 ROIs)", "hierarchy_early_visual_avg"))

            if 'hierarchy_late_semantic_avg' in self.data.columns:
                options.append(("Late Semantic Average (12 ROIs)", "hierarchy_late_semantic_avg"))

            if 'hierarchy_all_rois_avg' in self.data.columns:
                options.append(("All ROIs Average (19 ROIs)", "hierarchy_all_rois_avg"))

        # ========== ROI ENCODER DATA ==========
        options.append(("─────────── ROI ENCODER ───────────", "header_roi_encoder"))

        # SECTION 1: Averaged brain activation
        options.append(("AVERAGED BRAIN ACTIVATION", "header_roi_sim"))

        roi_sim_measures = [
            ('roi_cosine_common_avg_sim', 'Cosine - All Brain Regions'),
            ('roi_cosine_early_avg_sim', 'Cosine - Early Visual Regions'),
            ('roi_cosine_late_avg_sim', 'Cosine - Late Semantic Regions'),
            ('roi_pearson_common_avg_sim', 'Pearson - All Brain Regions'),
            ('roi_pearson_early_avg_sim', 'Pearson - Early Visual Regions'),
            ('roi_pearson_late_avg_sim', 'Pearson - Late Semantic Regions'),
        ]

        for col_name, display_name in roi_sim_measures:
            if col_name in self.data.columns:
                options.append((display_name, col_name))

        # SECTION 2: Brain activation patterns
        options.append(("BRAIN ACTIVATION PATTERNS", "header_roi_pattern"))

        roi_pattern_measures = [
            ('roi_cosine_common_avg_roi', 'Cosine - All Brain Regions'),
            ('roi_cosine_early_avg_roi', 'Cosine - Early Visual Regions'),
            ('roi_cosine_late_avg_roi', 'Cosine - Late Semantic Regions'),
            ('roi_pearson_common_avg_roi', 'Pearson - All Brain Regions'),
            ('roi_pearson_early_avg_roi', 'Pearson - Early Visual Regions'),
            ('roi_pearson_late_avg_roi', 'Pearson - Late Semantic Regions'),
        ]

        for col_name, display_name in roi_pattern_measures:
            if col_name in self.data.columns:
                options.append((display_name, col_name))

        # ========== VOXEL ENCODER DATA ==========
        if any(m.startswith('voxel_') and not m.startswith('voxel_to_roi_') for m in self.brain_measures):
            options.append(("─────────── VOXEL ENCODER ───────────", "header_voxel_encoder"))
            options.append(("VOXEL-LEVEL ANALYSIS", "header_voxel"))

            # Average voxel measures only (no subject-level)
            voxel_measures = [
                ('voxel_cosine_all_avg', 'Cosine - All Voxels'),
                ('voxel_cosine_early_all_avg', 'Cosine - Early Region Voxels'),
                ('voxel_cosine_late_all_avg', 'Cosine - Late Region Voxels'),
                ('voxel_pearson_all_avg', 'Pearson - All Voxels'),
                ('voxel_pearson_early_all_avg', 'Pearson - Early Region Voxels'),
                ('voxel_pearson_late_all_avg', 'Pearson - Late Region Voxels'),
            ]

            for col_name, display_name in voxel_measures:
                if col_name in self.data.columns:
                    options.append((display_name, col_name))

        # ========== INDIVIDUAL ROI REGIONS (VOXEL PATTERNS) ==========
        roi_voxel_pearson_cols = [col for col in self.brain_measures if col.startswith('roi_voxel_pearson_')]
        roi_voxel_cosine_cols = [col for col in self.brain_measures if col.startswith('roi_voxel_cosine_')]

        if roi_voxel_pearson_cols or roi_voxel_cosine_cols:
            options.append(("─────────── INDIVIDUAL ROI REGIONS (Voxel Patterns) ───────────", "header_roi_voxel"))

            # Define region categories for organization
            early_regions = ['V1d', 'V1v', 'V2d', 'V2v', 'V3d', 'V3v', 'hV4']
            face_object_regions = ['EBA', 'FBA2', 'OFA', 'FFA1', 'FFA2']
            scene_semantic_regions = ['OPA', 'PPA', 'RSC', 'OWFA', 'VWFA1', 'VWFA2', 'mfswords']

            # Pearson section
            if roi_voxel_pearson_cols:
                options.append(("PEARSON CORRELATION (Voxel Patterns)", "header_roi_voxel_pearson"))

                # Early visual regions
                options.append(("  Early Visual", "header_roi_voxel_pearson_early"))
                for region in early_regions:
                    col_name = f'roi_voxel_pearson_{region}'
                    if col_name in roi_voxel_pearson_cols:
                        options.append((f"  {region}", col_name))

                # Face/Object regions
                options.append(("  Face & Object", "header_roi_voxel_pearson_face"))
                for region in face_object_regions:
                    col_name = f'roi_voxel_pearson_{region}'
                    if col_name in roi_voxel_pearson_cols:
                        options.append((f"  {region}", col_name))

                # Scene/Semantic regions
                options.append(("  Scene & Semantic", "header_roi_voxel_pearson_scene"))
                for region in scene_semantic_regions:
                    col_name = f'roi_voxel_pearson_{region}'
                    if col_name in roi_voxel_pearson_cols:
                        options.append((f"  {region}", col_name))

            # Cosine section
            if roi_voxel_cosine_cols:
                options.append(("COSINE SIMILARITY (Voxel Patterns)", "header_roi_voxel_cosine"))

                # Early visual regions
                options.append(("  Early Visual", "header_roi_voxel_cosine_early"))
                for region in early_regions:
                    col_name = f'roi_voxel_cosine_{region}'
                    if col_name in roi_voxel_cosine_cols:
                        options.append((f"  {region}", col_name))

                # Face/Object regions
                options.append(("  Face & Object", "header_roi_voxel_cosine_face"))
                for region in face_object_regions:
                    col_name = f'roi_voxel_cosine_{region}'
                    if col_name in roi_voxel_cosine_cols:
                        options.append((f"  {region}", col_name))

                # Scene/Semantic regions
                options.append(("  Scene & Semantic", "header_roi_voxel_cosine_scene"))
                for region in scene_semantic_regions:
                    col_name = f'roi_voxel_cosine_{region}'
                    if col_name in roi_voxel_cosine_cols:
                        options.append((f"  {region}", col_name))

        return options