Update Chest_Xray_Report_Generator-V2.py

Chest_Xray_Report_Generator-V2.py

@@ -1,307 +1,307 @@
-import os
-import transformers
-from transformers import pipeline
-import gradio as gr
-import cv2
-import numpy as np
-import pydicom
-
-##### Libraries For Grad-Cam-View
-import os
-import cv2
-import numpy as np
-import torch
-from functools import partial
-from torchvision import transforms
-from pytorch_grad_cam import GradCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, EigenGradCAM, LayerCAM, FullGrad
-from pytorch_grad_cam.utils.image import show_cam_on_image, preprocess_image
-from pytorch_grad_cam.ablation_layer import AblationLayerVit
-from transformers import VisionEncoderDecoderModel
-
-def generate_gradcam(image_path, model_path, output_path, method='gradcam', use_cuda=True, aug_smooth=False, eigen_smooth=False):
-    methods = {
-        "gradcam": GradCAM,
-        "scorecam": ScoreCAM,
-        "gradcam++": GradCAMPlusPlus,
-        "ablationcam": AblationCAM,
-        "xgradcam": XGradCAM,
-        "eigencam": EigenCAM,
-        "eigengradcam": EigenGradCAM,
-        "layercam": LayerCAM,
-        "fullgrad": FullGrad
-    }
-
-    if method not in methods:
-        raise ValueError(f"Method should be one of {list(methods.keys())}")
-
-    model = VisionEncoderDecoderModel.from_pretrained(model_path)
-    model.encoder.eval()
-
-    if use_cuda and torch.cuda.is_available():
-        model.encoder = model.encoder.cuda()
-    else:
-        use_cuda = False
-
-    #target_layers = [model.blocks[-1].norm1]  ## For ViT model
-    #target_layers = model.blocks[-1].norm1    ## For EfficientNet-B7 model
-    target_layers = [model.encoder.encoder.layer[-1].layernorm_before]  ## For ViT-based VisionEncoderDecoder model
-    #target_layers = [model.encoder.encoder.layers[-1].blocks[-1].layernorm_before, model.encoder.encoder.layers[-1].blocks[0].layernorm_before] ## For Swin-based VisionEncoderDecoder mode
-    
-
-    if method == "ablationcam":
-        cam = methods[method](model=model.encoder,
-                              target_layers=target_layers,
-                              use_cuda=use_cuda,
-                              reshape_transform=reshape_transform,
-                              ablation_layer=AblationLayerVit())
-    else:
-        cam = methods[method](model=model.encoder,
-                              target_layers=target_layers,
-                              use_cuda=use_cuda,
-                              reshape_transform=reshape_transform)
-
-    rgb_img = cv2.imread(image_path, 1)[:, :, ::-1]
-    rgb_img = cv2.resize(rgb_img, (224, 224)) ## (224, 224)
-    rgb_img = np.float32(rgb_img) / 255
-    input_tensor = preprocess_image(rgb_img, mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
-
-    targets = None
-    cam.batch_size = 16
-
-    grayscale_cam = cam(input_tensor=input_tensor, targets=targets, eigen_smooth=eigen_smooth, aug_smooth=aug_smooth)
-    grayscale_cam = grayscale_cam[0, :]
-
-    cam_image = show_cam_on_image(rgb_img, grayscale_cam)
-    output_file = os.path.join(output_path, 'gradcam_result.png')
-    cv2.imwrite(output_file, cam_image)
-
-
-def reshape_transform(tensor, height=14, width=14):  ### height=14, width=14 for ViT-based Model
-    batch_size, token_number, embed_dim = tensor.size()
-    if token_number < height * width:
-        pad = torch.zeros(batch_size, height * width - token_number, embed_dim, device=tensor.device)
-        tensor = torch.cat([tensor, pad], dim=1)
-    elif token_number > height * width:
-        tensor = tensor[:, :height * width, :]
-
-    result = tensor.reshape(batch_size, height, width, embed_dim)
-    result = result.transpose(2, 3).transpose(1, 2)
-    return result
-
-
-
-
-# Example usage:
-#image_path = "/home/chayan/CGI_Net/images/images/CXR1353_IM-0230-1001.png"
-model_path = "/home/chayan/ViT-GPT2/Mimic_test/"
-output_path = "/home/chayan/ViT-GPT2/CAM-Result/"
-
-
-
-def sentence_case(paragraph):
-    sentences = paragraph.split('. ')
-    formatted_sentences = [sentence.capitalize() for sentence in sentences if sentence]
-    formatted_paragraph = '. '.join(formatted_sentences)
-    return formatted_paragraph
-
-def dicom_to_png(dicom_file, png_file):
-    # Load DICOM file
-    dicom_data = pydicom.dcmread(dicom_file)
-    dicom_data.PhotometricInterpretation = 'MONOCHROME1'
-
-    # Normalize pixel values to 0-255
-    img = dicom_data.pixel_array
-    img = img.astype(np.float32)
-
-    img = cv2.normalize(img, None, 0, 255, cv2.NORM_MINMAX)
-    img = img.astype(np.uint8)
-
-    # Save as PNG
-    cv2.imwrite(png_file, img)
-    return img
-
-
-Image_Captioner = pipeline("image-to-text", model = "/home/chayan/ViT-GPT2/Mimic_test/")
-
-data_dir = '/home/chayan/ViT-GPT2/'
-
-def xray_report_generator(Image_file):
-  if Image_file[-4:] =='.dcm':
-    png_file = 'DCM2PNG.png'
-    dicom_to_png(Image_file, png_file)
-    Image_file = os.path.join(data_dir, png_file)
-    output = Image_Captioner(Image_file, max_new_tokens=512)
-
-  else:
-    output = Image_Captioner(Image_file, max_new_tokens=512)
-
-  result = output[0]['generated_text']
-  output_paragraph = sentence_case(result)
-  
-  generate_gradcam(Image_file, model_path, output_path, method='gradcam', use_cuda=True)
-  
-  grad_cam_image =  output_path + 'gradcam_result.png'
-
-  return Image_file,grad_cam_image, output_paragraph
-  
-
-
-def save_feedback(feedback):
-    feedback_dir = "/home/chayan/ViT-GPT2/Feedback/"  # Update this to your desired directory
-    if not os.path.exists(feedback_dir):
-        os.makedirs(feedback_dir)
-    feedback_file = os.path.join(feedback_dir, "feedback.txt")
-    with open(feedback_file, "a") as f:
-        f.write(feedback + "\n")
-    return "Feedback submitted successfully!"
-
-
-  
-  
-# Custom CSS styles
-custom_css = """
-<style>
-
-#title {
-    color: green;
-    font-size: 36px;
-    font-weight: bold;
-}
-#description {
-    color: green;
-    font-size: 22px;
-}
-
-
-#submit-btn {
-    background-color: #1E90FF; /* DodgerBlue */
-    color: green;
-    padding: 15px 32px;
-    text-align: center;
-    text-decoration: none;
-    display: inline-block;
-    font-size: 20px;
-    margin: 4px 2px;
-    cursor: pointer;
-}
-#submit-btn:hover {
-    background-color: #00FFFF;
-}
-
-.intext textarea {
-    color: green;
-    font-size: 20px;
-    font-weight: bold;
-}
-
-
-.small-button {
-    color: green;
-    padding: 5px 10px;
-    font-size: 20px;
-}
-
-</style>
-"""
-
-# Sample image paths
-sample_images = [
-    "/mnt/data/chayan/MIMIC-CXR-JPG/2.0.0/files/p19565388/s54621108/a9510716-02da91b0-61532c26-a65b2efc-c9dfa6f1.jpg",
-    "/mnt/data/chayan/MIMIC-CXR-JPG/2.0.0/files/p19454978/s52312858/93681764-ec39480e-0518b12c-199850c2-f15118ab.jpg",
-    "/mnt/data/chayan/MIMIC-CXR-JPG/2.0.0/files/p17340686/s55469953/6ff741e9-6ea01eef-1bf10153-d1b6beba-590b6620.jpg"
-    #"sample4.png",
-    #"sample5.png"
-]
-
-def set_input_image(image_path):
-    return gr.update(value=image_path)
-
-
-with gr.Blocks(css = custom_css) as demo:
-
-    #gr.HTML(custom_css)  # Inject custom CSS
-
-    gr.Markdown(
-         """
-        <h1 style="color:blue; font-size: 36px; font-weight: bold">Chest X-ray Report Generator</h1>
-        <p id="description">Upload an X-ray image and get its report with heat-map visualization.</p>
-        """
-    )
-
-    with gr.Row():
-        inputs = gr.File(label="Upload Chest X-ray Image File", type="filepath")
-    
-    with gr.Row():
-        with gr.Column(scale=1, min_width=300):
-            outputs1 = gr.Image(label="Image Viewer")
-        with gr.Column(scale=1, min_width=300):
-            outputs2 = gr.Image(label="Grad_CAM-Visualization")
-        with gr.Column(scale=1, min_width=300):
-            outputs3 = gr.Textbox(label="Generated Report", elem_classes = "intext")
-            
-
-    submit_btn = gr.Button("Generate Report", elem_id="submit-btn")
-    submit_btn.click(
-        fn=xray_report_generator,
-        inputs=inputs,
-        outputs=[outputs1, outputs2, outputs3])
-        
-    
-    gr.Markdown(
-        """
-        <h2 style="color:green; font-size: 24px;">Or choose a sample image:</h2>
-        """
-    )
-
-    with gr.Row():
-        for idx, sample_image in enumerate(sample_images):
-            with gr.Column(scale=1):
-                #sample_image_component = gr.Image(value=sample_image, interactive=False)
-                select_button = gr.Button(f"Select Sample Image {idx+1}")
-                select_button.click(
-                    fn=set_input_image,
-                    inputs=gr.State(value=sample_image),
-                    outputs=inputs
-                )
-
-
-
-      # Feedback section
-    gr.Markdown(
-        """
-        <h2 style="color:green; font-size: 24px;">Provide Your Valuable Feedback:</h2>
-        """
-    )
-
-    with gr.Row():
-        feedback_input = gr.Textbox(label="Your Feedback", lines=4, placeholder="Enter your feedback here...")
-        feedback_submit_btn = gr.Button("Submit Feedback", elem_classes="small-button")
-        feedback_output = gr.Textbox(label="Feedback Status", interactive=False)
-
-    feedback_submit_btn.click(
-        fn=save_feedback,
-        inputs=feedback_input,
-        outputs=feedback_output
-    )
-   
-   
-   
-demo.launch(share=True)
-
-
-# inputs = gr.File(label="Upload Chest X-ray Image File", type="filepath")
-# outputs1 =gr.Image(label="Image Viewer")
-# outputs2 =gr.Image(label="Grad_CAM-Visualization")
-# outputs3 = gr.Textbox(label="Generated Report")
-
-
-# interface = gr.Interface(
-#     fn=xray_report_generator,
-#     inputs=inputs,
-#     outputs=[outputs1, outputs2, outputs3],
-#     title="Chest X-ray Report Generator",
-#     description="Upload an X-ray image and get its report.",
-# )
-
-
+import os
+import transformers
+from transformers import pipeline
+import gradio as gr
+import cv2
+import numpy as np
+import pydicom
+
+##### Libraries For Grad-Cam-View
+import os
+import cv2
+import numpy as np
+import torch
+from functools import partial
+from torchvision import transforms
+from pytorch_grad_cam import GradCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, EigenGradCAM, LayerCAM, FullGrad
+from pytorch_grad_cam.utils.image import show_cam_on_image, preprocess_image
+from pytorch_grad_cam.ablation_layer import AblationLayerVit
+from transformers import VisionEncoderDecoderModel
+
+def generate_gradcam(image_path, model_path, output_path, method='gradcam', use_cuda=True, aug_smooth=False, eigen_smooth=False):
+    methods = {
+        "gradcam": GradCAM,
+        "scorecam": ScoreCAM,
+        "gradcam++": GradCAMPlusPlus,
+        "ablationcam": AblationCAM,
+        "xgradcam": XGradCAM,
+        "eigencam": EigenCAM,
+        "eigengradcam": EigenGradCAM,
+        "layercam": LayerCAM,
+        "fullgrad": FullGrad
+    }
+
+    if method not in methods:
+        raise ValueError(f"Method should be one of {list(methods.keys())}")
+
+    model = VisionEncoderDecoderModel.from_pretrained(model_path)
+    model.encoder.eval()
+
+    if use_cuda and torch.cuda.is_available():
+        model.encoder = model.encoder.cuda()
+    else:
+        use_cuda = False
+
+    #target_layers = [model.blocks[-1].norm1]  ## For ViT model
+    #target_layers = model.blocks[-1].norm1    ## For EfficientNet-B7 model
+    target_layers = [model.encoder.encoder.layer[-1].layernorm_before]  ## For ViT-based VisionEncoderDecoder model
+    #target_layers = [model.encoder.encoder.layers[-1].blocks[-1].layernorm_before, model.encoder.encoder.layers[-1].blocks[0].layernorm_before] ## For Swin-based VisionEncoderDecoder mode
+    
+
+    if method == "ablationcam":
+        cam = methods[method](model=model.encoder,
+                              target_layers=target_layers,
+                              use_cuda=use_cuda,
+                              reshape_transform=reshape_transform,
+                              ablation_layer=AblationLayerVit())
+    else:
+        cam = methods[method](model=model.encoder,
+                              target_layers=target_layers,
+                              use_cuda=use_cuda,
+                              reshape_transform=reshape_transform)
+
+    rgb_img = cv2.imread(image_path, 1)[:, :, ::-1]
+    rgb_img = cv2.resize(rgb_img, (224, 224)) ## (224, 224)
+    rgb_img = np.float32(rgb_img) / 255
+    input_tensor = preprocess_image(rgb_img, mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+
+    targets = None
+    cam.batch_size = 16
+
+    grayscale_cam = cam(input_tensor=input_tensor, targets=targets, eigen_smooth=eigen_smooth, aug_smooth=aug_smooth)
+    grayscale_cam = grayscale_cam[0, :]
+
+    cam_image = show_cam_on_image(rgb_img, grayscale_cam)
+    output_file = os.path.join(output_path, 'gradcam_result.png')
+    cv2.imwrite(output_file, cam_image)
+
+
+def reshape_transform(tensor, height=14, width=14):  ### height=14, width=14 for ViT-based Model
+    batch_size, token_number, embed_dim = tensor.size()
+    if token_number < height * width:
+        pad = torch.zeros(batch_size, height * width - token_number, embed_dim, device=tensor.device)
+        tensor = torch.cat([tensor, pad], dim=1)
+    elif token_number > height * width:
+        tensor = tensor[:, :height * width, :]
+
+    result = tensor.reshape(batch_size, height, width, embed_dim)
+    result = result.transpose(2, 3).transpose(1, 2)
+    return result
+
+
+
+
+# Example usage:
+#image_path = "/home/chayan/CGI_Net/images/images/CXR1353_IM-0230-1001.png"
+model_path = "./Mimic_test/"
+output_path = "./CAM-Result/"
+
+
+
+def sentence_case(paragraph):
+    sentences = paragraph.split('. ')
+    formatted_sentences = [sentence.capitalize() for sentence in sentences if sentence]
+    formatted_paragraph = '. '.join(formatted_sentences)
+    return formatted_paragraph
+
+def dicom_to_png(dicom_file, png_file):
+    # Load DICOM file
+    dicom_data = pydicom.dcmread(dicom_file)
+    dicom_data.PhotometricInterpretation = 'MONOCHROME1'
+
+    # Normalize pixel values to 0-255
+    img = dicom_data.pixel_array
+    img = img.astype(np.float32)
+
+    img = cv2.normalize(img, None, 0, 255, cv2.NORM_MINMAX)
+    img = img.astype(np.uint8)
+
+    # Save as PNG
+    cv2.imwrite(png_file, img)
+    return img
+
+
+Image_Captioner = pipeline("image-to-text", model = "./Mimic_test/")
+
+data_dir = output_path
+
+def xray_report_generator(Image_file):
+  if Image_file[-4:] =='.dcm':
+    png_file = 'DCM2PNG.png'
+    dicom_to_png(Image_file, png_file)
+    Image_file = os.path.join(data_dir, png_file)
+    output = Image_Captioner(Image_file, max_new_tokens=512)
+
+  else:
+    output = Image_Captioner(Image_file, max_new_tokens=512)
+
+  result = output[0]['generated_text']
+  output_paragraph = sentence_case(result)
+  
+  generate_gradcam(Image_file, model_path, output_path, method='gradcam', use_cuda=True)
+  
+  grad_cam_image =  output_path + 'gradcam_result.png'
+
+  return Image_file,grad_cam_image, output_paragraph
+  
+
+
+def save_feedback(feedback):
+    feedback_dir = "./Feedback/"  # Update this to your desired directory
+    if not os.path.exists(feedback_dir):
+        os.makedirs(feedback_dir)
+    feedback_file = os.path.join(feedback_dir, "feedback.txt")
+    with open(feedback_file, "a") as f:
+        f.write(feedback + "\n")
+    return "Feedback submitted successfully!"
+
+
+  
+  
+# Custom CSS styles
+custom_css = """
+<style>
+
+#title {
+    color: green;
+    font-size: 36px;
+    font-weight: bold;
+}
+#description {
+    color: green;
+    font-size: 22px;
+}
+
+
+#submit-btn {
+    background-color: #1E90FF; /* DodgerBlue */
+    color: green;
+    padding: 15px 32px;
+    text-align: center;
+    text-decoration: none;
+    display: inline-block;
+    font-size: 20px;
+    margin: 4px 2px;
+    cursor: pointer;
+}
+#submit-btn:hover {
+    background-color: #00FFFF;
+}
+
+.intext textarea {
+    color: green;
+    font-size: 20px;
+    font-weight: bold;
+}
+
+
+.small-button {
+    color: green;
+    padding: 5px 10px;
+    font-size: 20px;
+}
+
+</style>
+"""
+
+# Sample image paths
+sample_images = [
+    "./Test-Images/p19565388/s54621108/a9510716-02da91b0-61532c26-a65b2efc-c9dfa6f1.jpg",
+    "./Test-Images/93681764-ec39480e-0518b12c-199850c2-f15118ab.jpg",
+    "./Test-Images/6ff741e9-6ea01eef-1bf10153-d1b6beba-590b6620.jpg"
+    #"sample4.png",
+    #"sample5.png"
+]
+
+def set_input_image(image_path):
+    return gr.update(value=image_path)
+
+
+with gr.Blocks(css = custom_css) as demo:
+
+    #gr.HTML(custom_css)  # Inject custom CSS
+
+    gr.Markdown(
+         """
+        <h1 style="color:blue; font-size: 36px; font-weight: bold">Chest X-ray Report Generator</h1>
+        <p id="description">Upload an X-ray image and get its report with heat-map visualization.</p>
+        """
+    )
+
+    with gr.Row():
+        inputs = gr.File(label="Upload Chest X-ray Image File", type="filepath")
+    
+    with gr.Row():
+        with gr.Column(scale=1, min_width=300):
+            outputs1 = gr.Image(label="Image Viewer")
+        with gr.Column(scale=1, min_width=300):
+            outputs2 = gr.Image(label="Grad_CAM-Visualization")
+        with gr.Column(scale=1, min_width=300):
+            outputs3 = gr.Textbox(label="Generated Report", elem_classes = "intext")
+            
+
+    submit_btn = gr.Button("Generate Report", elem_id="submit-btn")
+    submit_btn.click(
+        fn=xray_report_generator,
+        inputs=inputs,
+        outputs=[outputs1, outputs2, outputs3])
+        
+    
+    gr.Markdown(
+        """
+        <h2 style="color:green; font-size: 24px;">Or choose a sample image:</h2>
+        """
+    )
+
+    with gr.Row():
+        for idx, sample_image in enumerate(sample_images):
+            with gr.Column(scale=1):
+                #sample_image_component = gr.Image(value=sample_image, interactive=False)
+                select_button = gr.Button(f"Select Sample Image {idx+1}")
+                select_button.click(
+                    fn=set_input_image,
+                    inputs=gr.State(value=sample_image),
+                    outputs=inputs
+                )
+
+
+
+      # Feedback section
+    gr.Markdown(
+        """
+        <h2 style="color:green; font-size: 24px;">Provide Your Valuable Feedback:</h2>
+        """
+    )
+
+    with gr.Row():
+        feedback_input = gr.Textbox(label="Your Feedback", lines=4, placeholder="Enter your feedback here...")
+        feedback_submit_btn = gr.Button("Submit Feedback", elem_classes="small-button")
+        feedback_output = gr.Textbox(label="Feedback Status", interactive=False)
+
+    feedback_submit_btn.click(
+        fn=save_feedback,
+        inputs=feedback_input,
+        outputs=feedback_output
+    )
+   
+   
+   
+demo.launch(share=True)
+
+
+# inputs = gr.File(label="Upload Chest X-ray Image File", type="filepath")
+# outputs1 =gr.Image(label="Image Viewer")
+# outputs2 =gr.Image(label="Grad_CAM-Visualization")
+# outputs3 = gr.Textbox(label="Generated Report")
+
+
+# interface = gr.Interface(
+#     fn=xray_report_generator,
+#     inputs=inputs,
+#     outputs=[outputs1, outputs2, outputs3],
+#     title="Chest X-ray Report Generator",
+#     description="Upload an X-ray image and get its report.",
+# )
+
+
 # interface.launch(share=True)