AI for Pediatric Medicine

Module: Healthcare AI | Difficulty: Advanced

Growth Percentile

Head Circumference Z-Score

Pediatric AI Challenges

import torch
import torch.nn as nn
import numpy as np

class AgeAwareClassifier(nn.Module):
    def __init__(self, num_classes=10, max_age=18):
        super().__init__()
        self.age_embedding = nn.Embedding(max_age + 1, 32)
        self.image_encoder = models.resnet34(pretrained=True)
        num_features = self.image_encoder.fc.in_features
        self.image_encoder.fc = nn.Identity()
        self.classifier = nn.Sequential(
            nn.Linear(num_features + 32, 128), nn.ReLU(),
            nn.Dropout(0.3), nn.Linear(128, num_classes))

    def forward(self, x, age):
        image_features = self.image_encoder(x)
        age_emb = self.age_embedding(age)
        combined = torch.cat([image_features, age_emb], dim=1)
        return self.classifier(combined)

class GrowthChartPredictor(nn.Module):
    def __init__(self):
        super().__init__()
        self.encoder = nn.Sequential(
            nn.Linear(4, 64), nn.ReLU(), nn.Linear(64, 32), nn.ReLU())
        self.predictor = nn.Linear(32, 3)

    def forward(self, age, sex, weight, height):
        features = torch.cat([age.unsqueeze(1), sex.unsqueeze(1),
                             weight.unsqueeze(1), height.unsqueeze(1)], dim=1)
        encoded = self.encoder(features)
        return self.predictor(encoded)

def compute_z_score(value, mean, std):
    return (value - mean) / (std + 1e-8)

model = AgeAwareClassifier(num_classes=10)
x = torch.randn(1, 3, 224, 224)
age = torch.tensor([5])
output = model(x, age)
print(f'Pediatric classification output: {output.shape}')

growth_model = GrowthChartPredictor()
percentiles = growth_model(
    torch.tensor([5.0]), torch.tensor([1]),
    torch.tensor([18.0]), torch.tensor([110.0]))
print(f'Growth percentiles: {percentiles.detach().numpy()[0]}')

Research Insight: Pediatric AI models must account for normal developmental changes: brain MRI appearance, bone density, and organ sizes vary dramatically with age. Transfer learning from adult models often fails because pediatric anatomy is fundamentally different. Federated learning across children's hospitals can address the data scarcity problem while maintaining privacy.