Rydhi commited on
Commit
9b9dc60
·
1 Parent(s): 52822e3

Remove standalone FLAIR panel, keep 3-region breakdown only

Browse files
Files changed (2) hide show
  1. app.py +29 -33
  2. inference.py +17 -15
app.py CHANGED
@@ -1,7 +1,6 @@
1
  import spaces
2
  import gradio as gr
3
  import torch
4
- import numpy as np
5
  from model import MedSegNet
6
  from inference import preprocess_volume, run_inference, make_slice_figure, make_combined_figure
7
  from report import build_pdf_report
@@ -29,15 +28,14 @@ model.eval()
29
 
30
 
31
  @spaces.GPU
32
- def run(t1_file, t1ce_file, t2_file, flair_file):
33
- if not all([t1_file, t1ce_file, t2_file, flair_file]):
34
- raise gr.Error("Please upload all 4 modalities: T1, T1ce, T2, and FLAIR.")
35
 
36
  model.to("cuda")
37
  volume, affine = preprocess_volume(t1_file, t1ce_file, t2_file, flair_file)
38
  pred = run_inference(model, volume, "cuda")
39
 
40
- # simple volumetric stats (voxel counts -> approx cm^3 assuming 1mm^3 voxels post-crop)
41
  tc_voxels = int(pred[0].sum())
42
  wt_voxels = int(pred[1].sum())
43
  et_voxels = int(pred[2].sum())
@@ -58,10 +56,10 @@ def run(t1_file, t1ce_file, t2_file, flair_file):
58
 
59
  def load_example():
60
  return (
 
61
  "examples/BRATS_001_0001.nii.gz", # T1
62
  "examples/BRATS_001_0002.nii.gz", # T1ce
63
  "examples/BRATS_001_0003.nii.gz", # T2
64
- "examples/BRATS_001_0000.nii.gz", # FLAIR
65
  )
66
 
67
 
@@ -73,58 +71,56 @@ theme = gr.themes.Soft(
73
  )
74
 
75
  with gr.Blocks(title="MedSegNet — 3D Brain Tumor Segmentation", theme=theme) as demo:
76
- gr.Markdown(
77
- """
78
- # 🧠 MedSegNet
79
- ### Transformer-based 3D Brain Tumor Segmentation from Multi-Modal MRI
80
-
81
- Final Year Project Demonstration — SegFormer3D architecture trained on the BraTS2017 dataset.
82
- Upload four co-registered MRI modalities (T1, T1ce, T2, FLAIR) to generate an automated
83
- tumor segmentation across three clinically relevant regions.
84
- """
85
- )
86
 
87
  with gr.Row():
88
  with gr.Column(scale=1):
89
  gr.Markdown("#### 1. Upload MRI Volumes")
90
  gr.Markdown("*(.nii or .nii.gz format, all 4 modalities required)*")
 
91
  t1 = gr.File(label="T1")
92
  t1ce = gr.File(label="T1ce")
93
  t2 = gr.File(label="T2")
 
 
 
 
 
 
 
 
 
94
  flair = gr.File(label="FLAIR")
 
 
 
95
 
96
  with gr.Row():
97
  example_btn = gr.Button("📁 Load Example Case", variant="secondary")
98
  submit_btn = gr.Button("▶ Run Segmentation", variant="primary")
99
 
100
- gr.Markdown(
101
- """
102
- ---
103
- #### About
104
- **Architecture:** SegFormer3D (hierarchical transformer encoder + lightweight MLP decoder)
105
- **Training data:** BraTS2017
106
- **Output regions:**
107
- - 🟡 Whole Tumor (WT) — total abnormal tissue
108
- - 🟢 Tumor Core (TC) — solid tumor mass
109
- - 🔴 Enhancing Tumor (ET) — actively enhancing region
110
-
111
- *Research/demonstration output only — not for clinical diagnosis.*
112
- """
113
- )
114
 
115
  with gr.Column(scale=2):
116
  gr.Markdown("#### 2. Segmentation Result")
117
  combined_image = gr.Image(label="Combined Segmentation Overlay")
118
  stats_output = gr.Markdown()
119
- pdf_output = gr.File(label="📄 Download Full PDF Report")
120
  gr.Markdown("#### Region Breakdown")
121
  output_image = gr.Image(label="FLAIR / WT / TC / ET")
122
 
123
- example_btn.click(fn=load_example, outputs=[t1, t1ce, t2, flair])
 
 
 
 
 
 
 
 
124
 
125
  submit_btn.click(
126
  fn=run,
127
- inputs=[t1, t1ce, t2, flair],
128
  outputs=[combined_image, output_image, stats_output, pdf_output],
129
  )
130
 
 
1
  import spaces
2
  import gradio as gr
3
  import torch
 
4
  from model import MedSegNet
5
  from inference import preprocess_volume, run_inference, make_slice_figure, make_combined_figure
6
  from report import build_pdf_report
 
28
 
29
 
30
  @spaces.GPU
31
+ def run(flair_file, t1_file, t1ce_file, t2_file):
32
+ if not all([flair_file, t1_file, t1ce_file, t2_file]):
33
+ raise gr.Error("Please upload all 4 modalities: FLAIR, T1, T1ce, and T2.")
34
 
35
  model.to("cuda")
36
  volume, affine = preprocess_volume(t1_file, t1ce_file, t2_file, flair_file)
37
  pred = run_inference(model, volume, "cuda")
38
 
 
39
  tc_voxels = int(pred[0].sum())
40
  wt_voxels = int(pred[1].sum())
41
  et_voxels = int(pred[2].sum())
 
56
 
57
  def load_example():
58
  return (
59
+ "examples/BRATS_001_0000.nii.gz", # FLAIR
60
  "examples/BRATS_001_0001.nii.gz", # T1
61
  "examples/BRATS_001_0002.nii.gz", # T1ce
62
  "examples/BRATS_001_0003.nii.gz", # T2
 
63
  )
64
 
65
 
 
71
  )
72
 
73
  with gr.Blocks(title="MedSegNet — 3D Brain Tumor Segmentation", theme=theme) as demo:
74
+ gr.Markdown("# MedSegNet")
 
 
 
 
 
 
 
 
 
75
 
76
  with gr.Row():
77
  with gr.Column(scale=1):
78
  gr.Markdown("#### 1. Upload MRI Volumes")
79
  gr.Markdown("*(.nii or .nii.gz format, all 4 modalities required)*")
80
+ flair = gr.File(label="FLAIR")
81
  t1 = gr.File(label="T1")
82
  t1ce = gr.File(label="T1ce")
83
  t2 = gr.File(label="T2")
84
+
85
+ with gr.Row():
86
+ example_btn = gr.Button("📁 Load Example Case", variant="secondary")
87
+ submit_btn = gr.Button("▶ Run Segmentation", variant="primary")
88
+
89
+
90
+ with gr.Column(scale=1):
91
+ gr.Markdown("#### 1. Upload MRI Volumes")
92
+ gr.Markdown("*(.nii or .nii.gz format, all 4 modalities required)*")
93
  flair = gr.File(label="FLAIR")
94
+ t1 = gr.File(label="T1")
95
+ t1ce = gr.File(label="T1ce")
96
+ t2 = gr.File(label="T2")
97
 
98
  with gr.Row():
99
  example_btn = gr.Button("📁 Load Example Case", variant="secondary")
100
  submit_btn = gr.Button("▶ Run Segmentation", variant="primary")
101
 
102
+ pdf_output = gr.DownloadButton("📄 Download PDF Report", variant="secondary")
 
 
 
 
 
 
 
 
 
 
 
 
 
103
 
104
  with gr.Column(scale=2):
105
  gr.Markdown("#### 2. Segmentation Result")
106
  combined_image = gr.Image(label="Combined Segmentation Overlay")
107
  stats_output = gr.Markdown()
 
108
  gr.Markdown("#### Region Breakdown")
109
  output_image = gr.Image(label="FLAIR / WT / TC / ET")
110
 
111
+ example_btn.click(fn=load_example, outputs=[flair, t1, t1ce, t2])
112
+
113
+ submit_btn.click(
114
+ fn=run,
115
+ inputs=[flair, t1, t1ce, t2],
116
+ outputs=[combined_image, output_image, stats_output, pdf_output],
117
+ )
118
+
119
+ example_btn.click(fn=load_example, outputs=[flair, t1, t1ce, t2])
120
 
121
  submit_btn.click(
122
  fn=run,
123
+ inputs=[flair, t1, t1ce, t2],
124
  outputs=[combined_image, output_image, stats_output, pdf_output],
125
  )
126
 
inference.py CHANGED
@@ -135,23 +135,25 @@ def make_slice_figure(volume, pred, save_path="preview.png"):
135
  return save_path
136
 
137
 
138
- def make_combined_figure(volume, pred, save_path="combined.png"):
139
  mid = find_best_slice(pred)
140
- flair_slice = volume[0][mid]
141
- pred_combined = combine_regions(pred)[mid]
142
-
143
- fig, ax = plt.subplots(figsize=(7, 7))
144
- ax.imshow(flair_slice, cmap="gray")
145
- ax.imshow(pred_combined, cmap=region_cmap, alpha=0.6, vmin=0, vmax=3)
146
- ax.set_title("Combined Prediction", fontsize=14)
147
- ax.axis("off")
148
-
149
- legend_elements = [
150
- Patch(facecolor="yellow", label="Whole Tumor (WT)"),
151
- Patch(facecolor="green", label="Tumor Core (TC)"),
152
- Patch(facecolor="red", label="Enhancing Tumor (ET)"),
153
  ]
154
- ax.legend(handles=legend_elements, loc="lower right", fontsize=9, framealpha=0.8)
 
 
 
 
 
 
 
155
 
156
  plt.tight_layout()
157
  plt.savefig(save_path, dpi=150, bbox_inches="tight", facecolor="white")
 
135
  return save_path
136
 
137
 
138
+ def make_slice_figure(volume, pred, save_path="preview.png"):
139
  mid = find_best_slice(pred)
140
+ flair_slice = volume[0][mid] # channel 0 = FLAIR
141
+
142
+ pred_tc, pred_wt, pred_et = pred[0][mid], pred[1][mid], pred[2][mid]
143
+
144
+ rows = [
145
+ ("Whole Tumor (WT)", pred_wt),
146
+ ("Tumor Core (TC)", pred_tc),
147
+ ("Enhancing Tumor (ET)", pred_et),
 
 
 
 
 
148
  ]
149
+
150
+ fig, axes = plt.subplots(1, 3, figsize=(15, 5))
151
+
152
+ for i, (region_name, region_slice) in enumerate(rows):
153
+ axes[i].imshow(flair_slice, cmap="gray")
154
+ axes[i].imshow(region_slice, cmap="Reds", alpha=0.5)
155
+ axes[i].set_title(region_name, fontsize=13)
156
+ axes[i].axis("off")
157
 
158
  plt.tight_layout()
159
  plt.savefig(save_path, dpi=150, bbox_inches="tight", facecolor="white")