breast_cancer_classifer.py

import os
import numpy as np
import tensorflow as tf
from tensorflow.keras.preprocessing.image import img_to_array, load_img
from huggingface_hub import hf_hub_download
import zipfile

# Step 1: Download the model directly from your Hugging Face repo
MODEL_REPO = "Ruchir456/breast-cancer-densenet"
MODEL_FILENAME = "best_model_savedmodel.zip"
EXTRACT_DIR = "best_model_savedmodel"

if not os.path.exists(EXTRACT_DIR):
    print("Downloading model from Hugging Face repo...")
    model_zip_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILENAME)
    print("✅ Model downloaded successfully!")

    # Extract the zip
    with zipfile.ZipFile(model_zip_path, 'r') as zip_ref:
        zip_ref.extractall(EXTRACT_DIR)
    print("✅ Model extracted successfully!")

# Step 2: Detect the actual SavedModel folder (with saved_model.pb inside)
saved_model_dir = None
for root, dirs, files in os.walk(EXTRACT_DIR):
    if "saved_model.pb" in files:
        saved_model_dir = root
        break

# Step 3: Load model once globally
try:
    if saved_model_dir:
        model = tf.keras.models.load_model(saved_model_dir, compile=False)
        print(f"✅ Model loaded successfully from {saved_model_dir}")
    else:
        model = None
        print("❌ Could not find saved_model.pb inside extracted folder.")
except Exception as e:
    model = None
    print(f"❌ Failed to load the model: {e}")

# Step 4: Internal mapping from model index to class label
class_labels = {0: 'benign', 1: 'malignant', 2: 'normal'}

# Step 5: Mapping from class label to detailed medical interpretation
class_descriptions = {
    'normal': (
        "*Normal:* Cells appear healthy and show no signs of cancer. "
        "There is no indication of abnormal tissue structure or cellular activity. "
        "No further investigation is typically required, but periodic screenings help ensure continued health."
    ),
    'benign': (
        "*Benign:* Cells may appear abnormal but are *non-cancerous* and do not spread to other tissues. "
        "These growths are usually slow-growing and not life-threatening. "
        "Monitoring for changes in size or behavior is common, and removal may be considered if symptoms develop."
    ),
    'malignant': (
        "*Malignant:* Cells are *cancerous*, showing uncontrolled growth, abnormal structure, and potential to invade nearby tissues or spread to distant organs. "
        "Further diagnostic evaluation and staging are important to determine the extent. "
        "Management may involve treatment plans such as surgery, chemotherapy, or radiation depending on the progression."
    )
}

# Step 6: Main prediction function
def breast_cancer_detection_model(image_path):
    try:
        if model is None:
            return "❌ Model is not loaded. Please check logs."

        # Load and preprocess the image
        img = load_img(image_path, target_size=(256, 256))
        img_array = img_to_array(img)
        img_array = np.expand_dims(img_array, axis=0) / 255.0

        # Make prediction
        prediction = model.predict(img_array)
        predicted_index = np.argmax(prediction)
        predicted_class = class_labels[predicted_index]
        confidence = prediction[0][predicted_index] * 100

        # Format and return interpretation
        result = class_descriptions[predicted_class]
        # Optional: add model confidence
        # result += f"\n📊 *Model Confidence:* {confidence:.2f}%"
        return result

    except Exception as e:
        return f"❌ Error processing the image: {str(e)}"