app.py

from PIL import Image, ImageOps
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd
from keras.models import load_model
import gradio as gr

# Load the model and class names outside the prediction function
model = load_model('keras_model.h5', compile=False)
class_names = [line.strip() for line in open('labels.txt', 'r')]

def create_plot(data):
    sns.set_theme(style="whitegrid")

    f, ax = plt.subplots(figsize=(5, 5))

    sns.set_color_codes("pastel")
    sns.barplot(x="Total", y="Labels", data=data, label="Total", color="b")

    sns.set_color_codes("muted")
    sns.barplot(x="Confidence Score", y="Labels", data=data, label="Conficence Score", color="b")

    ax.legend(ncol=2, loc="lower right", frameon=True)
    sns.despine(left=True, bottom=True)
    return f

def predict_tumor(img):
    np.set_printoptions(suppress=True)
    
    data = np.ndarray(shape=(1, 224, 224, 3), dtype=np.float32)

    # Resize and preprocess the input image
    size = (224, 224)
    image_PIL = Image.fromarray(img)
    image = ImageOps.fit(image_PIL, size, Image.LANCZOS)
    image_array = np.asarray(image)
    normalized_image_array = (image_array.astype(np.float32) / 127.0) - 1
    data[0] = normalized_image_array
    
    # Make a prediction
    prediction = model.predict(data)
    index = np.argmax(prediction)
    class_name = class_names[index]
    confidence_score = prediction[0][index]
    
    c_name = class_name.strip()
    tumor_prediction = f"Model {'detected' if c_name == 'Yes' else 'did not detect'} Tumor"
    other_class = 'No' if c_name == 'Yes' else 'Yes'
    
    # Prepare data for plotting
    res = {"Labels": [c_name, other_class], "Confidence Score": [(confidence_score * 100), (1 - confidence_score) * 100], "Total": 100}
    data_for_plot = pd.DataFrame.from_dict(res)

    tumor_conf_plt = create_plot(data_for_plot)
    
    return tumor_prediction, tumor_conf_plt


# Gradio Interface
with gr.Blocks(title="Brain Tumor Detection | Data Science Dojo", css="styles.css") as demo:
    with gr.Row():
        with gr.Column(scale=4):
            with gr.Row():
                imgInput = gr.Image()
        with gr.Column(scale=1):
            tumor = gr.Textbox(label='Presence of Tumor')
            plot = gr.Plot(label="Plot")
            
    submit_button = gr.Button(value="Submit")
    submit_button.click(fn=predict_tumor, inputs=[imgInput], outputs=[tumor, plot])

    gr.Examples(
        examples=["pred2.jpg", "pred3.jpg"],
        inputs=imgInput,
        outputs=[tumor, plot],
        fn=predict_tumor,
        cache_examples=True,
    )

demo.launch()