giuseppericcio commited on
Commit
dfcd20b
·
verified ·
1 Parent(s): 9a01dbb

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -5
README.md CHANGED
@@ -17,7 +17,7 @@ datasets:
17
  - OpenNeuro
18
  base_model:
19
  - google/medgemma-1.5-4b-it
20
- - google/medsiglip-base-patch16-448
21
  ---
22
 
23
  # 🧠 BrainGemma3D — Brain Report Automation via Inflated Vision Transformers in 3D
@@ -76,7 +76,6 @@ pip install torch torchvision transformers nibabel scikit-image lime
76
  ```python
77
  from huggingface_hub import snapshot_download
78
 
79
-
80
  # 1. Download the repository containing our custom architecture from Hugging Face
81
  repo_id = "praiselab-picuslab/BrainGemma3D"
82
  print(f"Downloading repository: {repo_id}...")
@@ -98,7 +97,6 @@ from medgemma3d_architecture import MedGemma3D, load_nifti_volume, CANONICAL_PRO
98
  device = "cuda" if torch.cuda.is_available() else "cpu"
99
  print(f"Hardware accelerator selected: {device}")
100
 
101
-
102
  # 2. Instantiate the base architecture (3D-inflated MedSigLIP + MedGemma)
103
  model = MedGemma3D(
104
  vision_model_dir=f"{local_dir}/vision_model",
@@ -110,7 +108,6 @@ model = MedGemma3D(
110
  device_map={"": 0} if device == "cuda" else None,
111
  )
112
 
113
-
114
  # 3. Load projector
115
  proj_path = os.path.join(local_dir, "projector_vis_scale.pt")
116
  print(f"Loading custom projector weights from: {proj_path}...")
@@ -132,7 +129,6 @@ if ckpt.get("vis_scale") is not None:
132
  model.eval()
133
  print("✅ BrainGemma3D is fully loaded and ready for inference!")
134
 
135
-
136
  # 4. Load MRI scan
137
  volume = load_nifti_volume(
138
  "path/to/brain_flair.nii.gz",
@@ -256,6 +252,13 @@ weights, wvol = run_interpretability(
256
  - `lime_top_supervoxels_grid.png` — Most influential supervoxels
257
  - `lime_weights.json` — Supervoxel importance scores
258
 
 
 
 
 
 
 
 
259
  ---
260
 
261
  ## ⚙️ Model Details
 
17
  - OpenNeuro
18
  base_model:
19
  - google/medgemma-1.5-4b-it
20
+ - google/medsiglip-448
21
  ---
22
 
23
  # 🧠 BrainGemma3D — Brain Report Automation via Inflated Vision Transformers in 3D
 
76
  ```python
77
  from huggingface_hub import snapshot_download
78
 
 
79
  # 1. Download the repository containing our custom architecture from Hugging Face
80
  repo_id = "praiselab-picuslab/BrainGemma3D"
81
  print(f"Downloading repository: {repo_id}...")
 
97
  device = "cuda" if torch.cuda.is_available() else "cpu"
98
  print(f"Hardware accelerator selected: {device}")
99
 
 
100
  # 2. Instantiate the base architecture (3D-inflated MedSigLIP + MedGemma)
101
  model = MedGemma3D(
102
  vision_model_dir=f"{local_dir}/vision_model",
 
108
  device_map={"": 0} if device == "cuda" else None,
109
  )
110
 
 
111
  # 3. Load projector
112
  proj_path = os.path.join(local_dir, "projector_vis_scale.pt")
113
  print(f"Loading custom projector weights from: {proj_path}...")
 
129
  model.eval()
130
  print("✅ BrainGemma3D is fully loaded and ready for inference!")
131
 
 
132
  # 4. Load MRI scan
133
  volume = load_nifti_volume(
134
  "path/to/brain_flair.nii.gz",
 
252
  - `lime_top_supervoxels_grid.png` — Most influential supervoxels
253
  - `lime_weights.json` — Supervoxel importance scores
254
 
255
+ ### Expected Output Example
256
+
257
+ <div align="left">
258
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/662a12d70951c58269b066fb/UkQwmZRwkn-rlNlFBNVkH.png" alt="LIME Interpretability" width="50%">
259
+ <p><i>Figure 3: LIME attribution maps for BraTS training sample 072. Red regions show supervoxels that positively contribute to pathology predictions. The model correctly focuses on tumor-affected areas in the left parietal and frontal lobes.</i></p>
260
+ </div>
261
+
262
  ---
263
 
264
  ## ⚙️ Model Details