Pxl-Persona
AI2025-02-034 min read

White Paper: Amazon Bedrock for building AI health care assistants - The cost of memory, tokens, and hosting

Healthcare AI Use Case Breakdown with Amazon Bedrock (Llama 3.2) AI-powered applications in healthcare can provide symptom analysis, medical triage, patient Q&A, clini...

Healthcare AI Use Case Breakdown with Amazon Bedrock (Llama 3.2)

AI-powered applications in healthcare can provide symptom analysis, medical triage, patient Q&A, clinical documentation assistance, and real-time decision support. Below is a detailed breakdown of various healthcare scenarios, including cost estimates based on Llama 3.2 (11B) pricing on Amazon Bedrock.

1. AI-Powered Symptom Checker (Patient Chatbot)

Use Case:

A symptom checker chatbot that asks patients about their symptoms, provides potential causes, and recommends whether they should see a doctor.

Token Usage Assumptions:

  • Prompt (Medical context setup): 500 tokens

  • User Input Per Turn: 75 tokens

  • AI Response Per Turn: 150 tokens

  • RAG Retrieval (Medical Knowledge Database): 2,000 tokens every 5 turns

  • Total Turns: 15 (patient describes symptoms, AI asks clarifying questions, gives recommendations)

Turn | Base Tokens (User + AI) | RAG Tokens | Total Tokens Sent | Cumulative Tokens |

1 | 500 + 75 | 0 | 725 | 725 |

5 | 2200 + 75 | 2000 | 4275 | 10,125 |

10 | ~5000 + 75 | 2000 | 7075 | 21,300 |

15 | ~8500 + 75 | 2000 | 10,075 | 33,450 |

Estimated Cost:

  • Total tokens per session: ~33,450

  • Cost per 1,000 tokens: $0.00035

  • Cost per session: $0.0117

  • Cost for 1,000 patient assessments: $11.70

  • Additional RAG Costs: Dependent on medical knowledge sources like AWS HealthLake or external FHIR databases.

2. AI Assistant for Doctors (Clinical Documentation)

Use Case:

AI assists doctors by transcribing patient conversations, summarizing key findings, and drafting notes for Electronic Health Records (EHR).

Token Usage Assumptions:

  • Doctor-Patient Conversation (Transcription per turn): 300 tokens

  • AI Summary Output Per Turn: 400 tokens

  • Medical Data Retrieval (RAG): 3,000 tokens every 5 turns

  • Total Turns: 10 (Doctor asks about symptoms, AI drafts notes)

Turn | Transcription Tokens | AI Summary | RAG Tokens | Total Tokens Sent | Cumulative Tokens |

1 | 300 | 400 | 0 | 700 | 700 |

5 | 1500 | 2000 | 3000 | 6500 | 15,500 |

10 | 3000 | 4000 | 3000 | 10,000 | 32,500 |

Estimated Cost:

  • Total tokens per session: ~32,500

  • Cost per 1,000 tokens: $0.00035

  • Cost per session: $0.0114

  • Cost for 1,000 patient visits: $11.40

  • Additional Costs: AWS Transcribe (~$0.0004 per second of audio), AWS HealthLake for EHR integration.

3. AI for Medical Image Analysis (Doctor Consultation)

Use Case:

AI assists radiologists or pathologists by analyzing images (X-rays, MRIs, CT scans) and providing text-based insights.

Token Usage Assumptions:

  • Medical Image Analysis Summary (AI): 1,000 tokens

  • Doctor-Patient Explanation: 500 tokens per turn

  • AI Response Per Turn: 800 tokens

  • RAG (Medical Literature/Clinical Guidelines): 4,000 tokens per analysis

  • Total Turns: 5 (Doctor consults AI about findings)

Turn | Doctor Query | AI Response | RAG Tokens | Total Tokens Sent | Cumulative Tokens |

1 | 500 | 800 | 4000 | 5300 | 5300 |

3 | 1500 | 2400 | 4000 | 7900 | 17,100 |

5 | 2500 | 4000 | 4000 | 10,500 | 38,100 |

Estimated Cost:

  • Total tokens per session: ~38,100

  • Cost per 1,000 tokens: $0.00035

  • Cost per session: $0.0133

  • Cost for 1,000 analyses: $13.30

  • Additional Costs: AWS Medical Imaging for storing/analyzing X-rays, MRIs, etc.

4. Real-Time Patient Monitoring with AI Alerts

Use Case:

AI continuously analyzes vitals, symptoms, and medical history to provide alerts for anomalies in ICU or remote patient monitoring.

Token Usage Assumptions:

  • Vitals Data (Streaming input per cycle): 100 tokens

  • AI Interpretation: 200 tokens

  • RAG (Medical Reference or Alert Criteria): 2,000 tokens every 5 cycles

  • Total Cycles: 20 (AI monitors patient every 10 minutes)

Cycle | Sensor Data | AI Interpretation | RAG Tokens | Total Tokens Sent | Cumulative Tokens |

1 | 100 | 200 | 0 | 300 | 300 |

5 | 500 | 1000 | 2000 | 3500 | 7700 |

10 | 1000 | 2000 | 2000 | 5000 | 18,700 |

20 | 2000 | 4000 | 2000 | 8000 | 39,500 |

Estimated Cost:

  • Total tokens per session: ~39,500

  • Cost per 1,000 tokens: $0.00035

  • Cost per 24-hour monitoring: $0.0138

  • Cost for 1,000 patient-days: $13.80

  • Additional Costs: AWS IoT for medical device integration.

5. AI-Assisted Mental Health Chatbot

Use Case:

An AI-powered mental health assistant provides therapy, mood tracking, and crisis intervention.

Token Usage Assumptions:

  • Initial Prompt: 500 tokens

  • User Input Per Turn: 200 tokens

  • AI Response Per Turn: 300 tokens

  • RAG (Psychology Research & Coping Strategies): 3,000 tokens every 5 turns

  • Total Turns: 20

Turn | User Input | AI Response | RAG Tokens | Total Tokens Sent | Cumulative Tokens |

1 | 500 + 200 | 300 | 0 | 1000 | 1000 |

5 | 1800 | 3000 | 3000 | 7800 | 18,600 |

10 | 3800 | 6000 | 3000 | 12,800 | 39,700 |

20 | 7800 | 12,000 | 3000 | 22,800 | 82,500 |

Estimated Cost:

  • Total tokens per session: ~82,500

  • Cost per 1,000 tokens: $0.00035

  • Cost per therapy session: $0.0289

  • Cost for 1,000 therapy sessions: $28.90

  • Additional Costs: AWS Comprehend Medical for sentiment analysis.

Final Cost Comparison

Use Case | Cost per 1,000 Sessions |

Symptom Checker | $11.70 |

Doctor Assistant | $11.40 |

Medical Imaging | $13.30 |

Patient Monitoring | $13.80 |

Mental Health Chatbot | $28.90 |

How Our Avatars Actually Work

Behind the scenes of voice, streaming, and animation.

Back to resources