Ali78-d commited on
Commit
2ff01c6
·
verified ·
1 Parent(s): 44c2a44

Upload 6 files

Browse files
Medical-Image-Classification-ChestXray_Code.ipynb ADDED
The diff for this file is too large to render. See raw diff
 
plots/confusion_matrix.png ADDED
plots/training_performance.png ADDED
pneumonia_pipeline_kfold.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb9424ab8225e85da7552536b27460e00e45d71c5cd1a0847ee31e2beb3dcfa0
3
+ size 28442750
reports/classification_report.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ Pneumonia Prediction Model - Final Classification Report (K-Fold Averaged)
2
+ ======================================================================
3
+ precision recall f1-score support
4
+ NORMAL 0.953021 0.691453 0.801231 234.000000
5
+ PNEUMONIA 0.841135 0.979487 0.905022 390.000000
6
+ accuracy 0.871474 0.871474 0.871474 0.871474
7
+ macro avg 0.897078 0.835470 0.853127 624.000000
8
+ weighted avg 0.883092 0.871474 0.866100 624.000000
ui_components/gradio_interface.py ADDED
@@ -0,0 +1,98 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ # Required imports for the Gradio interface
3
+ import os, random, joblib, io
4
+ import numpy as np
5
+ import pandas as pd
6
+ import matplotlib.pyplot as plt
7
+ import seaborn as sns
8
+ from tqdm.notebook import tqdm
9
+ from PIL import Image
10
+
11
+ import torch
12
+ import torch.nn as nn
13
+ import torch.optim as optim
14
+ from torch.utils.data import DataLoader, Subset
15
+ from torchvision import datasets, transforms, models
16
+
17
+ import optuna
18
+ from pytorch_grad_cam import GradCAM
19
+ from pytorch_grad_cam.utils.image import show_cam_on_image
20
+ from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
21
+
22
+ import gradio as gr
23
+ from sklearn.model_selection import StratifiedKFold
24
+ from sklearn.metrics import classification_report, confusion_matrix
25
+
26
+ # Define device, IMG_SIZE, and get_model function if not globally available
27
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
28
+ IMG_SIZE = (224, 224) # Assuming this is the image size used for preprocessing
29
+
30
+ def get_model(model_name, pretrained=True):
31
+ weights = 'DEFAULT' if pretrained else None
32
+ if model_name == "resnet50":
33
+ model = models.resnet50(weights=weights)
34
+ model.fc = nn.Linear(model.fc.in_features, 2)
35
+ elif model_name == "efficientnet_b0":
36
+ model = models.efficientnet_b0(weights=weights)
37
+ model.classifier[1] = nn.Linear(model.classifier[1].in_features, 2)
38
+ elif model_name == "densenet121":
39
+ model = models.densenet121(weights=weights)
40
+ model.classifier = nn.Linear(model.classifier.in_features, 2)
41
+ else:
42
+ raise ValueError("Model name not recognized")
43
+ return model
44
+
45
+ # Define the path where your model pipeline is saved
46
+ MODEL_PIPELINE_PATH = "/content/drive/MyDrive/pneumonia_pipeline_kfold.pth"
47
+
48
+ # 1. تحميل الأنبوب المتكامل من المسار المحدد
49
+ pipeline = torch.load(MODEL_PIPELINE_PATH, map_location=device, weights_only=False)
50
+
51
+ # 2. استخراج المكونات من الأنبوب
52
+ model_name = pipeline['model_name']
53
+ class_names = pipeline['class_names']
54
+ preprocess = pipeline['transforms']
55
+
56
+ # 3. بناء الموديل وتحميل الأوزان
57
+ model = get_model(model_name, pretrained=False).to(device)
58
+ model.load_state_dict(pipeline['model_state_dict'])
59
+ model.eval()
60
+
61
+ # 4. تحديد طبقة الـ CAM تلقائياً
62
+ if model_name == "resnet50":
63
+ target_layers = [model.layer4[-1]]
64
+ elif model_name == "efficientnet_b0":
65
+ target_layers = [model.features[-1][0]]
66
+ elif model_name == "densenet121":
67
+ target_layers = [model.features.norm5]
68
+
69
+ cam = GradCAM(model=model, target_layers=target_layers)
70
+
71
+ def predict_and_explain(image):
72
+ if image is None: return None, None
73
+ img_rgb = image.convert("RGB")
74
+ input_tensor = preprocess(img_rgb).unsqueeze(0).to(device)
75
+ with torch.no_grad():
76
+ output = model(input_tensor)
77
+ probs = torch.softmax(output, dim=1)[0]
78
+ confidences = {class_names[i]: float(probs[i]) for i in range(len(class_names))}
79
+ grayscale_cam = cam(input_tensor=input_tensor, targets=[ClassifierOutputTarget(probs.argmax().item())], aug_smooth=True, eigen_smooth=True)[0, :]
80
+ img_np = np.array(img_rgb.resize(IMG_SIZE)) / 255.0
81
+ cam_image = show_cam_on_image(img_np, grayscale_cam, use_rgb=True)
82
+ return confidences, cam_image
83
+
84
+ # إطلاق واجهة Gradio مع إضافة أسماء الفريق في الوصف
85
+ gr.Interface(
86
+ fn=predict_and_explain,
87
+ inputs=gr.Image(type="pil", label="ارفع صورة الأشعة هنا"),
88
+ outputs=[gr.Label(label="التشخيص"), gr.Image(label="التفسير البصري")],
89
+ title=f"نظام تشخيص الالتهاب الرئوي ({model_name})",
90
+ description="""
91
+ ### 🦅 تم تطوير وتدريب هذا النموذج بواسطة فريق النسور:
92
+ * **المهندس عبدالله الجحافي**
93
+ * **المهندس أيمن الوهبي**
94
+ * **المهندس واثق المريسي**
95
+ * **المهندس علي المحرابي**
96
+ """,
97
+ theme=gr.themes.Soft()
98
+ ).launch(debug=True, share=True)