AI-Generated Dental Treatment Plans: System for Clinics

How AI Automates the Dental Treatment Plan?

A doctor spends 15–25 minutes on a treatment plan: analyzing images, examination data, history, forming a procedure sequence with ICD and ICD-C codes, calculating cost, and creating a patient document. Coding errors lead to insurance denials – up to 3–4 cases per month in large networks. We have developed an AI system that takes over documentation and generates a structured plan based on the doctor's data. This is not a substitute for diagnosis but an assistant that reduces time to 6 minutes and boosts coding accuracy to 94%. The savings per doctor are calculated individually and depend on the clinic's scope.

Why AI Reduces Insurance Denials?

Manual plan creation is tedious, wasting doctor time and creating error risks. The AI system not only speeds up the process but also structures data according to ICD-10-CM and SNODENT (see Wikipedia). Compare:

Result: doctor saves 70% time, insurance accepts the plan first time.

What Data is Needed for Plan Generation?

After an exam, the doctor inputs or dictates: tooth chart, identified pathologies per tooth, patient priorities. The system generates:

• Full treatment plan with procedure sequence
• ICD-10-CM and ICD-C codes for insurance
• Alternative treatment options (conservative vs radical)
• Cost breakdown by stage
• Informed consent for the patient (in plain language)

from langchain_openai import ChatOpenAI
from pydantic import BaseModel
from typing import Optional
import json

class ToothCondition(BaseModel):
    tooth_number: int  # по ISO 3950
    diagnosis: str
    severity: str      # mild / moderate / severe
    priority: str      # urgent / planned / cosmetic

class TreatmentPlan(BaseModel):
    patient_id: str
    chief_complaint: str
    diagnoses: list[ToothCondition]
    treatment_phases: list[dict]   # [{phase, procedures, duration_weeks, cost_range}]
    total_visits_estimate: int
    contraindications: list[str]
    alternative_options: list[dict]
    informed_consent_summary: str

class DentalTreatmentPlanGenerator:
    SYSTEM_PROMPT = """Ты — AI-ассистент стоматолога. Помогаешь формализовать план лечения.
Ты НЕ ставишь диагноз — ты структурируешь данные, предоставленные врачом.
Используй актуальные стандарты: МКБ-10-СМ, МКБ-С (SNODENT), СанПиН 2.1.3.2630-10.
Последовательность процедур должна соответствовать клинической логике:
сначала неотложная помощь → гигиенические процедуры → терапия → хирургия → ортопедия."""

    def __init__(self):
        self.llm = ChatOpenAI(model="gpt-4o", temperature=0.1)

    def generate_plan(
        self,
        patient_data: dict,
        tooth_conditions: list[ToothCondition],
        patient_preferences: dict
    ) -> TreatmentPlan:
        conditions_text = "\n".join([
            f"Зуб {tc.tooth_number}: {tc.diagnosis} ({tc.severity}), приоритет: {tc.priority}"
            for tc in tooth_conditions
        ])

        prompt = f"""Создай план стоматологического лечения.

Данные пациента:
- Возраст: {patient_data.get('age')}
- Аллергии: {patient_data.get('allergies', 'не указаны')}
- Системные заболевания: {patient_data.get('systemic_conditions', 'нет')}
- Принимаемые препараты: {patient_data.get('medications', 'нет')}
- Главная жалоба: {patient_data.get('chief_complaint')}

Состояние зубов (по данным врача):
{conditions_text}

Предпочтения пациента:
- Бюджет: {patient_preferences.get('budget', 'не ограничен')}
- Приоритет: {patient_preferences.get('priority', 'качество')} (качество/скорость/бюджет)
- Страховка: {patient_preferences.get('insurance', 'нет')}

Создай план с:
1. Этапы лечения (фазы с обоснованием последовательности)
2. Для каждой процедуры: название, код МКБ-С, количество посещений, риски
3. Альтернативный план (более консервативный)
4. Предупреждения и противопоказания
5. Краткое резюме для пациента (без медицинского жаргона)

Верни JSON структуры TreatmentPlan."""

        response = self.llm.invoke([
            {"role": "system", "content": self.SYSTEM_PROMPT},
            {"role": "user", "content": prompt}
        ])

        return TreatmentPlan.model_validate_json(response.content)

How AI Analyzes X-Ray Images?

For clinics with digital X-rays—data extraction via Vision API. The model describes changes on panoramic images: carious cavities, periapical changes, bone loss. The result is used as auxiliary information for the doctor.

import base64
from openai import OpenAI

client = OpenAI()

def analyze_dental_xray(image_path: str) -> dict:
    """Анализирует рентгеновский снимок — вспомогательно для врача"""
    with open(image_path, "rb") as f:
        image_b64 = base64.b64encode(f.read()).decode()

    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "user",
            "content": [
                {"type": "image_url",
                 "image_url": {"url": f"data:image/jpeg;base64,{image_b64}"}},
                {"type": "text",
                 "text": """Опиши видимые изменения на панорамном рентгеновском снимке зубов.
Структурируй по зонам. Укажи: кариозные полости, периапикальные изменения,
потеря костной ткани, состояние корневых каналов.
ВАЖНО: Это вспомогательная информация для врача, не диагноз."""}
            ]
        }],
        max_tokens=500
    )
    return {"xray_observations": response.choices[0].message.content}

How Does It Integrate with MIS?

We connect to 1C:Meditsina, Dental4Windows, Ident. The plan is loaded as scheduled procedures linked to the patient, with status "planned" and ICD-C code.

# Интеграция с 1С:Медицина, Dental4Windows, Ident
class DentalMISConnector:
    def push_treatment_plan(self, plan: TreatmentPlan, mis_patient_id: str):
        """Загружает план в медицинскую информационную систему"""
        procedures = []
        for phase in plan.treatment_phases:
            for proc in phase["procedures"]:
                procedures.append({
                    "code": proc["icds_code"],
                    "name": proc["name"],
                    "tooth_number": proc.get("tooth_number"),
                    "phase": phase["phase_number"],
                    "estimated_cost": proc.get("cost_range"),
                    "status": "planned"
                })

        return self.mis_client.create_treatment_plan(
            patient_id=mis_patient_id,
            procedures=procedures,
            created_by="ai_assistant"
        )

What's Included in the Work

The project includes:

• Basic plan generator on GPT-4o with custom prompt and validation pipeline
• Integration with MIS (1C:Meditsina, Dental4Windows, Ident) via REST API
• X-ray analysis module via Vision API
• Data schema configuration for your standards and encoders
• 2-day training for doctors and administrators
• 3 months of technical support

Our Experience and Results

We have implemented over 10 AI projects in healthcare, including for a network of 8 dental clinics. The team has 5+ years in MLOps and NLP. We guarantee coding accuracy of at least 90% at start (94% based on our data). We use certified OpenAI models and our own fine-tuned models for clinic specifics.

Case study: a network of 8 dental clinics. Average time to create a treatment plan: 22 min → 6 min (doctor reviews and corrects AI draft). ICD-C code match accuracy (checked by insurance department): 94%. First 4 months: 0 insurance denials due to incorrect coding (previously 3–4 per month).

Timeline: basic plan generator: 3–4 weeks; integration with MIS and X-ray analysis: an additional 6–8 weeks.

Contact us for a demo. Request a pilot project—we will evaluate your infrastructure and prepare a proposal.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.