ResNet-18-pneumonia-detection
| Property | Value |
|---|---|
| Model Type | Classification (Binary) |
| Architecture | ResNet-18 |
| Accuracy | 80.4% |
| Input Size | 224x224 RGB |
| Model URL | HuggingFace |
What is ResNet-18-pneumonia-detection?
ResNet-18-pneumonia-detection is a specialized deep learning model designed for automated pneumonia detection from chest X-ray images. Built on the ResNet-18 architecture and fine-tuned on the Chest X-ray Pneumonia Dataset from Kaggle, this model provides binary classification between normal and pneumonia cases with optimized performance through FP16 quantization.
Implementation Details
The model leverages the robust ResNet-18 architecture, modified with a custom final linear layer (512 to 2 output features) for binary classification. It processes 224x224 RGB images and applies standard normalization techniques. The implementation includes FP16 quantization for improved inference efficiency while maintaining high accuracy.
- Pre-processing includes image resizing and normalization with mean=[0.485, 0.456, 0.406] and std=[0.229, 0.224, 0.225]
- Training utilized Adam optimizer with 1e-4 learning rate over 10 epochs
- Batch size of 16 with Cross-Entropy loss function
- FP16 quantization for optimized deployment
Core Capabilities
- Binary classification of chest X-rays (Normal vs. Pneumonia)
- Fast inference with FP16 optimization
- Precision: 78.2%
- Recall: 75.8%
- F1-Score: 79.5%
Frequently Asked Questions
Q: What makes this model unique?
This model combines the proven ResNet-18 architecture with FP16 quantization, offering an optimal balance between accuracy and inference speed for pneumonia detection. Its specialized training on chest X-rays makes it particularly effective for clinical screening applications.
Q: What are the recommended use cases?
The model is designed for preliminary screening of chest X-rays in clinical settings. However, it should only be used as a supporting tool, not as a primary diagnostic method. All results should be verified by qualified radiologists due to the possibility of false positives or negatives.
