oral-cancer-classifer

Oral Cancer Classification Model

Model ID

ankon1/oral-cancer-classifer

Model Description

A comprehensive deep learning system for classifying oral cancer images from clinical and histopathological samples. Combines multiple state-of-the-art architectures including Vision Transformers (ViT) and convolutional neural networks (ResNet50, DenseNet121, InceptionV3, VGG16) to detect cancerous tissue in oral cavity images. Processes both clinical photographs and microscopic histopathological images, achieving 95.7% accuracy on independent test data with built-in explainability features using LIME and SHAP.

Uses

Primary Use: Classification of oral tissue images as cancerous or non-cancerous for research and educational purposes

Research Applications:

• Medical imaging research
• Dataset generation for rare oral cancer subtypes
• Transfer learning for other medical imaging tasks

Educational Tool: Assisting dental/medical students in oral lesion identification

Screening Assistance: Preliminary screening in telemedicine applications (requires clinical validation)

Important: This model is NOT for direct clinical diagnosis without FDA/regulatory approval and should not replace professional medical consultation.

How to Get Started with the Model

# Install required packages
pip install transformers torch tensorflow pillow numpy

# Basic inference example
from transformers import AutoModelForImageClassification, AutoFeatureExtractor
import torch
from PIL import Image

# Load model and feature extractor
model = AutoModelForImageClassification.from_pretrained(
    "[Your-Username]/oral-cancer-classifier",
    trust_remote_code=True
)
feature_extractor = AutoFeatureExtractor.from_pretrained(
    "[Your-Username]/oral-cancer-classifier"
)

# Load and preprocess image
image = Image.open("oral_tissue_sample.jpg").convert("RGB")
inputs = feature_extractor(images=image, return_tensors="pt")

# Make prediction
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits
    predicted_class = torch.argmax(logits, dim=1).item()
    
# Interpret results
if predicted_class == 1:
    print("⚠️ Cancer detected - Please consult a healthcare professional")
else:
    print("✅ Non-cancerous tissue detected")
    
# Get confidence scores
probabilities = torch.nn.functional.softmax(logits, dim=1)
print(f"Cancer: {probabilities[0][1]:.2%}, Non-Cancer: {probabilities[0][0]:.2%}")

Quick Test with Sample Image:

# Test with a sample image URL
import requests
from io import BytesIO

# Download sample oral image
url = "https://example.com/sample_oral_image.jpg"
response = requests.get(url)
image = Image.open(BytesIO(response.content))

# Process and predict
inputs = feature_extractor(images=image, return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs)
    prediction = torch.argmax(outputs.logits, dim=1).item()
print(f"Prediction: {'Cancer' if prediction == 1 else 'Non-Cancer'}")

Limitations

1. Clinical Validation: Not validated for direct clinical diagnosis; requires FDA/regulatory approval before clinical use
2. Data Limitations: Trained on limited dataset (500 cancer, 250 non-cancer images); may not generalize to all populations
3. Image Quality Dependency: Performance significantly drops with low-quality, blurry, or poorly lit images
4. Early Detection: Reduced accuracy for detecting early-stage/pre-cancerous lesions

Future Improvements

1. Data Expansion: Incorporate larger, more diverse datasets from multiple geographic regions
2. Multi-modal Integration: Fuse imaging data with patient metadata (age, smoking history, genetic factors)
3. 3D Imaging Support: Extend to 3D volumetric data from CT/MRI scans
4. Federated Learning: Enable privacy-preserving training across multiple institutions
5. Real-time Optimization: Develop lightweight versions for mobile/edge device deployment