Use cases

Real healthcare work, as training data.

From clinical reasoning and documentation to speech, imaging and care over time, see how AI teams use Mozu data to train and evaluate models on real healthcare work.

Clinical reasoning

Data for models that need to follow a patient encounter from first question to final decision.

From conversation to diagnosis and treatment

Train models on doctor patient conversation transcripts paired with what the doctor decided next: the diagnosis, the prescription, the referral and the follow-up. Models learn the documented clinical reasoning that connects what a patient says to what a doctor does.

Transcripts: 100,000; Transcripts

Data · Doctor Patient Conversations

Outcome prediction

Where follow-up data is available, teams can study how earlier decisions relate to later events recorded in care.

Data · Doctor Patient Conversations

Clinical documentation and administrative work

Data for models and agents that write and act inside patient records.

Visit notes from conversations

Pair each conversation with the note the doctor actually wrote, giving models a target for summarizing a real visit into a clinical note.

Average turns per conversation: 42; Average turns per conversation
Average words per conversation: 666; Average words per conversation

Data · Doctor Patient Conversations

Patient record tasks

Tasks grounded in real encounters, such as entering orders, prescriptions and referrals, and the administrative steps around a visit, for teams building agents that complete work in patient records.

Data · Doctor Patient Conversations

Medical speech and transcription

Data for speech models where clinical details matter.

Medical transcription and speaker attribution

Train and evaluate speech models on real medical conversations with transcripts, speaker diarization and grading by medical doctors.

Hours of Japanese medical audio: 12,500; Hours of Japanese medical audio

Data · Japanese Medical Audio

Catching critical errors

Doctor-graded transcripts with extra review of medications, dosages, symptoms and diagnoses: the details where a single transcription error changes the meaning of a visit.

Data · Japanese Medical Audio

Care over time

Data for models that need to reason across visits, not one encounter at a time.

Longitudinal psychiatric care

Repeated visits with the same patients, from a detailed intake through regular check-ins, so models can learn how symptoms change and why doctors adjust medication and dosage.

Data · Psychiatry Longitudinal Encounters

Follow-up across encounters

Patients who return for more than one visit let models connect an earlier decision to what the patient reports next.

Data · Doctor Patient Conversations

Patient records and images of conditions

Data for models that combine images, conversations and records.

Treatment response in dermatology

Before and after images linked to diagnosis, medication, dosage and transcripts of the patient’s visits, so models can connect what a doctor sees and hears to the treatment chosen and assess changes after treatment.

Data · Dermatology Before/After Images

Evaluations and benchmarks

Tests built around meaningful clinical work, for finding where models fall short.

Clinical evaluations

Use encounter context, documented decisions and follow-up records to test whether models can handle meaningful clinical tasks.

Benchmarks in development

Our benchmark work covers dermatology treatment response and mental health across time.

The two research datasets behind many of these use cases have public summaries in the dataset index.

Browse research datasets

Tell us what your models need.

Request dataset samples