Automatic Medical Dictation Transcription Implementation

Implementation of automatic transcription of medical dictations Medical dictation - the doctor dictates the examination, anamnesis, diagnosis, prescriptions - the system creates structured text in the required medical record format. Saves the doctor's time: 30-60 minutes per day. Specifics: high cost of error, medical terminology, abbreviations, Latin names. ### System requirements - WER <5% for medical terms - Structuring in SOAP/HL7 FHIR format - HIPAA/152-FZ compatibility - data in a secure environment - Integration with MIS (1C: Medicine, Medialog, EMIAS) ### Medical dictation architecture

from enum import Enum
from dataclasses import dataclass

class MedicalSection(Enum):
    COMPLAINT = "complaint"       # Жалобы
    ANAMNESIS = "anamnesis"      # Анамнез
    OBJECTIVE = "objective"       # Объективный осмотр
    DIAGNOSIS = "diagnosis"       # Диагноз
    TREATMENT = "treatment"       # Назначения

@dataclass
class MedicalRecord:
    patient_id: str
    doctor_id: str
    sections: dict[MedicalSection, str]
    raw_transcript: str
    created_at: str

class MedicalDictationProcessor:
    def __init__(self):
        # Whisper дообученный на медицинских данных
        self.stt = WhisperModel(
            "whisper-medical-ru-v1",
            device="cuda",
            compute_type="float16"
        )
        self.medical_normalizer = MedicalTextNormalizer()

    async def process_dictation(
        self,
        audio_path: str,
        patient_context: dict
    ) -> MedicalRecord:
        # 1. Транскрибируем с медицинским словарём
        segments, _ = self.stt.transcribe(
            audio_path,
            language="ru",
            initial_prompt="Медицинская диктовка врача. Жалобы, анамнез, диагноз, назначения."
        )
        raw_text = " ".join(seg.text for seg in segments)

        # 2. Нормализация медицинской лексики
        normalized = self.medical_normalizer.normalize(raw_text)

        # 3. Структурирование через LLM
        structured = await self.structure_medical_text(normalized, patient_context)

        return MedicalRecord(
            patient_id=patient_context["patient_id"],
            doctor_id=patient_context["doctor_id"],
            sections=structured,
            raw_transcript=raw_text,
            created_at=datetime.utcnow().isoformat()
        )

    async def structure_medical_text(self, text: str, context: dict) -> dict:
        response = await client.chat.completions.create(
            model="gpt-4o",
            messages=[{
                "role": "system",
                "content": """Ты медицинский редактор. Структурируй диктовку врача.
                Разбей на разделы: Жалобы, Анамнез болезни, Объективный осмотр,
                Диагноз (МКБ-10 код), Назначения.
                Исправь медицинские термины. JSON ответ."""
            }, {
                "role": "user",
                "content": f"Пациент: {context.get('age')} лет, {context.get('gender')}.\n{text}"
            }],
            response_format={"type": "json_object"}
        )
        return json.loads(response.choices[0].message.content)
```### Medical abbreviations```python
MEDICAL_ABBREVIATIONS = {
    "ад": "артериальное давление",
    "чсс": "частота сердечных сокращений",
    "жкт": "желудочно-кишечный тракт",
    "орви": "острая респираторная вирусная инфекция",
    # Расшифровываем при диктовке, сокращаем в финальном тексте
}
```### Features of working with the HIS Integration via the FHIR R4 API or specific APIs of 1C:Medicine, Medialog. We create ClinicalDocument or DocumentReference in FHIR. ### Implementation timeframes - Basic system: 4–6 weeks - Additional training of Whisper on medical data: +4–6 weeks (if a dataset is available) - Integration with a specific HIS: +2–4 weeks