| import gradio as gr | |
| import numpy as np | |
| from PIL import Image | |
| from keras.models import Model | |
| from keras.layers import Input, Conv2D, MaxPooling2D, Conv2DTranspose, concatenate | |
| def preprocess_image(image, size): | |
| image = image.resize((size, size)) | |
| image = image.convert("L") | |
| image = np.array(image) / 255.0 | |
| return image | |
| def conv_block(input, num_filters): | |
| conv = Conv2D(num_filters, (3, 3), activation="relu", padding="same", kernel_initializer='he_normal')(input) | |
| conv = Conv2D(num_filters, (3, 3), activation="relu", padding="same", kernel_initializer='he_normal')(conv) | |
| return conv | |
| def encoder_block(input, num_filters): | |
| conv = conv_block(input, num_filters) | |
| pool = MaxPooling2D((2, 2))(conv) | |
| return conv, pool | |
| def decoder_block(input, skip_features, num_filters): | |
| uconv = Conv2DTranspose(num_filters, (2, 2), strides=2, padding="same")(input) | |
| con = concatenate([uconv, skip_features]) | |
| conv = conv_block(con, num_filters) | |
| return conv | |
| def build_model(input_shape, num_output_channels): | |
| input_layer = Input(input_shape) | |
| s1, p1 = encoder_block(input_layer, 64) | |
| s2, p2 = encoder_block(p1, 128) | |
| s3, p3 = encoder_block(p2, 256) | |
| s4, p4 = encoder_block(p3, 512) | |
| b1 = conv_block(p4, 1024) | |
| d1 = decoder_block(b1, s4, 512) | |
| d2 = decoder_block(d1, s3, 256) | |
| d3 = decoder_block(d2, s2, 128) | |
| d4 = decoder_block(d3, s1, 64) | |
| output_layer = Conv2D(num_output_channels, 1, padding="same", activation="sigmoid")(d4) | |
| model = Model(input_layer, output_layer, name="U-Net") | |
| model.load_weights('modelo.h5') | |
| return model | |
| def image_segmentation(image, size=128, num_output_channels=1): | |
| image = preprocess_image(image, size) | |
| image = np.expand_dims(image, 0) | |
| output = model.predict(image, verbose=0) | |
| mask_image = output[0] | |
| mask_image = np.squeeze(mask_image, -1) | |
| mask_image *= 255 | |
| mask_image = mask_image.astype(np.uint8) | |
| mask_image = Image.fromarray(mask_image).convert("L") | |
| positive_pixels = np.count_nonzero(mask_image) | |
| total_pixels = mask_image.size[0] * mask_image.size[1] | |
| percentage = (positive_pixels / total_pixels) * 100 | |
| class_0_percentage = 100 - percentage | |
| class_1_percentage = percentage | |
| return mask_image, class_0_percentage, class_1_percentage | |
| if __name__ == "__main__": | |
| size = 128 | |
| num_output_channels = 1 | |
| model = build_model(input_shape=(size, size, 1), num_output_channels=num_output_channels) | |
| gr.Interface( | |
| fn=image_segmentation, | |
| inputs="image", | |
| outputs=[ | |
| gr.Image(type="pil", label="Cáncer de mama "), | |
| gr.Number(label="Benigno"), | |
| gr.Number(label="Maligno") | |
| ], | |
| title='<h1 style="text-align: center;">Cáncer de mama </h1>', | |
| description=""" | |
| Demostración de segmentación de imágenes de ultrasonido de cáncer de mama. | |
| """, | |
| theme="default", | |
| layout="vertical", | |
| verbose=True | |
| ).launch(debug=True) | |