STREAMLITE / test_model.py
Stroke-ia's picture
Create test_model.py
6f78e96 verified
Raw History Blame Contribute Delete
3.14 kB
import os
import sys
import torch
import nibabel as nib
import numpy as np
from pathlib import Path
import SimpleITK as sitk
# =======================
# CONFIG
# =======================
SAVE_DIR = "./results"
os.makedirs(SAVE_DIR, exist_ok=True)
MODEL_PATH = "model.best" # ton modèle entraîné
# =======================
# CHARGEMENT DU MODÈLE
# =======================
def load_model():
print("🔄 Chargement du modèle...")
model = torch.load(MODEL_PATH, map_location="cpu")
model.eval()
return model
# =======================
# PRÉPROCESSING
# =======================
def load_image(file_path):
ext = Path(file_path).suffix.lower()
if ext in [".nii", ".gz"]:
img = nib.load(file_path)
data = img.get_fdata()
affine = img.affine
return data, affine
elif ext == ".dcm":
itk_img = sitk.ReadImage(file_path)
data = sitk.GetArrayFromImage(itk_img)
affine = np.eye(4) # placeholder simple
return data, affine
else:
raise ValueError("❌ Format non supporté (seulement NIfTI ou DICOM).")
# =======================
# INFERENCE
# =======================
def run_inference(model, data):
data = (data - np.min(data)) / (np.max(data) - np.min(data) + 1e-8)
data_tensor = torch.tensor(data, dtype=torch.float32).unsqueeze(0).unsqueeze(0) # (B,C,H,W,D)
with torch.no_grad():
pred = model(data_tensor)
pred = torch.argmax(pred, dim=1).squeeze().cpu().numpy()
return pred
# =======================
# RAPPORT
# =======================
def generate_report(pred, save_path):
volume_total = np.prod(pred.shape)
volume_tumeur = np.sum(pred > 0)
ratio = (volume_tumeur / volume_total) * 100
report = [
"📄 RAPPORT MÉDICAL AUTOMATISÉ",
"-----------------------------------",
f"Dimensions de l'image : {pred.shape}",
f"Volume total voxels : {volume_total}",
f"Volume suspect tumeur : {volume_tumeur}",
f"Ratio suspect : {ratio:.2f} %",
"",
"⚠️ Rapport automatique à valider par un médecin."
]
with open(save_path, "w", encoding="utf-8") as f:
f.write("\n".join(report))
print("\n".join(report))
# =======================
# MAIN
# =======================
if __name__ == "__main__":
if len(sys.argv) < 2:
print("❌ Utilisation : python test_model.py <chemin_fichier_IRM>")
sys.exit(1)
input_path = sys.argv[1]
print(f"📂 Fichier en entrée : {input_path}")
# Load
model = load_model()
data, affine = load_image(input_path)
# Inference
pred = run_inference(model, data)
# Save segmentation
out_seg_path = os.path.join(SAVE_DIR, "segmentation_pred.nii.gz")
nib.save(nib.Nifti1Image(pred.astype(np.uint8), affine), out_seg_path)
print(f"✅ Segmentation sauvegardée : {out_seg_path}")
# Save report
out_report_path = os.path.join(SAVE_DIR, "segmentation_report_detailed.txt")
generate_report(pred, out_report_path)
print(f"✅ Rapport sauvegardé : {out_report_path}")