AI for Rare Disease Diagnosis and Research

Module: Healthcare AI | Difficulty: Advanced

Few-Shot Learning (Prototypical)

where .

Knowledge Graph Embedding

Rare Disease Statistics

import torch
import torch.nn as nn

class PrototypicalNetwork(nn.Module):
    def __init__(self, input_dim=100, hidden_dim=64, embedding_dim=32):
        super().__init__()
        self.encoder = nn.Sequential(
            nn.Linear(input_dim, hidden_dim), nn.ReLU(),
            nn.Linear(hidden_dim, hidden_dim), nn.ReLU(),
            nn.Linear(hidden_dim, embedding_dim))

    def forward(self, support_set, query):
        support_embeddings = self.encoder(support_set)
        query_embedding = self.encoder(query)
        prototypes = support_embeddings.mean(dim=1)
        distances = torch.cdist(query_embedding, prototypes)
        return -distances

class RareDiseaseClassifier(nn.Module):
    def __init__(self, num_symptoms=200, num_diseases=500):
        super().__init__()
        self.symptom_embed = nn.Embedding(num_symptoms, 32)
        self.disease_embed = nn.Embedding(num_diseases, 32)
        self.gating = nn.Sequential(
            nn.Linear(32, 1), nn.Sigmoid())

    def forward(self, symptom_ids, disease_ids):
        symptom_emb = self.symptom_embed(symptom_ids).mean(dim=1)
        disease_emb = self.disease_embed(disease_ids)
        gate = self.gating(symptom_emb.unsqueeze(1) * disease_emb)
        scores = (symptom_emb.unsqueeze(1) * disease_emb * gate).sum(dim=-1)
        return scores

proto_net = PrototypicalNetwork(input_dim=200, embedding_dim=32)
support = torch.randn(5, 10, 200)
query = torch.randn(1, 200)
distances = proto_net(support, query)
print(f'Distances to prototypes: {distances.shape}')

disease_classifier = RareDiseaseClassifier(num_symptoms=200, num_diseases=500)
symptoms = torch.randint(0, 200, (1, 15))
diseases = torch.randint(0, 500, (1, 100))
scores = disease_classifier(symptoms, diseases)
print(f'Disease similarity scores: {scores.shape}')

Research Insight: Rare disease diagnosis is one of the most impactful applications of AI in healthcare. Few-shot learning approaches that can recognize diseases from just a handful of examples are particularly valuable. Knowledge graphs that connect genes, symptoms, and diseases provide a structured representation that can generalize to unseen diseases through relational reasoning.