Medical Imaging with Deep Learning

Why Medical Imaging AI?

Radiologists process thousands of images daily, and the volume grows approximately 10% per year. Deep learning models achieve radiologist-level performance across imaging modalities.

• Reduce interpretation time by 30-50%
• Catch subtle findings missed during high-volume reads
• Provide consistent performance regardless of fatigue
• Enable screening in under-resourced settings

Convolutional Neural Networks (CNNs)

A CNN learns hierarchical features through convolutional, pooling, and fully connected layers.

Convolution Operation

Batch Normalization

ResNet Skip Connections

import torch
import torch.nn as nn
import torchvision.models as models

class MedicalResNet(nn.Module):
    def __init__(self, num_classes=2, pretrained=True):
        super().__init__()
        self.backbone = models.resnet50(pretrained=pretrained)
        in_features = self.backbone.fc.in_features
        self.backbone.fc = nn.Sequential(
            nn.Dropout(0.5),
            nn.Linear(in_features, 256),
            nn.ReLU(),
            nn.Dropout(0.3),
            nn.Linear(256, num_classes)
        )

    def forward(self, x):
        return self.backbone(x)

Data Augmentation for Limited Medical Datasets

Geometric Augmentations

Intensity Augmentations

• Contrast adjustment:
• Gaussian noise:

from torchvision import transforms

medical_transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.RandomHorizontalFlip(p=0.5),
    transforms.RandomRotation(15),
    transforms.ColorJitter(brightness=0.2, contrast=0.2),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

Mixup

where .

Transfer Learning Strategies

1. Feature extraction: Freeze pretrained layers
2. Gradual unfreezing: Progressively unfreeze layers
3. Full fine-tuning: Update all weights with small learning rate

def create_medical_model(num_classes):
    model = models.resnet50(pretrained=False)
    checkpoint = torch.load('medical_resnet50.pth')
    model.load_state_dict(checkpoint['model_state_dict'])
    model.fc = nn.Linear(model.fc.in_features, num_classes)
    return model

Evaluation Metrics

Practical Considerations

• Class imbalance: Use focal loss or class-weighted cross-entropy
• Data privacy: Consider federated learning for multi-site studies
• Explainability: Use Grad-CAM to visualize model attention