README.md

metadata

title: Vertebro
emoji: 🦴
colorFrom: yellow
colorTo: purple
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: true
hf_oauth: true
hf_oauth_scopes:
  - inference-api
short_description: AI-driven spine billing and operative note analyzer

🦴 Vertebro — Automated Spine CPT Billing & Operative Note NLP

Live Demo: Vertebro (Hugging Face Space)
Developed by: Shubham Laiwala (Nazarian Lab, BIDMC)
Advisor: Dr. Jason Pittman, Dr. Ara Nazarian Version: v1.0 — CPT Logic Core: deterministic spine billing inference

---

💡 Overview

Vertebro is an intelligent, rule-based NLP engine that automatically interprets spine operative notes and suggests the most relevant CPT billing codes, modifiers, and rationales — all while maintaining payer-aware logic (Medicare, BCBS, Aetna, etc.).

It’s built for:

• Research: Rapid auditing of spine operative notes
• Billing validation: Cross-checking documentation against CPT standards
• Clinical education: Understanding how procedural phrasing drives billing logic

The underlying logic is deterministic, explainable, and optimized for cervical, thoracic, and lumbar spine cases.

---

🧠 Key Features

Feature	Description
CPT Suggestion Engine	Extracts primary and add-on CPT codes from free-text operative notes
Payer Logic	Adjusts code and modifier recommendations for each payer
Multi-Region Support	Recognizes cervical, thoracic, and lumbar levels
Explainable Outputs	Every CPT code includes rationale and confidence score
Modifier Awareness	Detects 22, 50, 58, 78, 79, 82, and laterality codes automatically
No PHI Retention	Inputs are transient and never stored or logged

---

🧩 How to Use Vertebro

1. Open the app: Vertebro on Hugging Face
2. Paste or type an operative note (e.g., “Left-sided TLIF L4–L5 with pedicle screws and interbody cage; posterolateral fusion performed. Navigation used.”)
3. Select the payer (e.g., Medicare, BCBS, Aetna)
4. Click “Submit.”
5. Review results:
   • CPT suggestions
   • Descriptions
   • Rationales
   • Categories (fusion, instrumentation, graft, etc.)
   • Confidence levels

💬 Please feel free to test real-life, de-identified operative note examples from your own cases.
The more diverse the phrasing, the better we can evaluate the model’s robustness.

---

🔍 Example Operative Notes to Try

1. Anterior Cervical Discectomy and Fusion (ACDF)

Anterior cervical discectomy and fusion performed at C5–C6 and C6–C7. PEEK cages placed at both levels with anterior plate fixation.

---

2. Transforaminal Lumbar Interbody Fusion (TLIF)

Left-sided TLIF L4–L5 with pedicle screws and interbody cage. Posterolateral fusion performed with local autograft.

---

3. Posterior Cervical Fusion with Instrumentation

Posterior cervical fusion from C3–C5 with lateral mass screws and rods. Local autograft used for fusion.

---

4. Hardware Removal

Removal of posterior instrumentation at L2–L4. No reinstrumentation performed.

---

5. Decompression Only

L3–L4 decompression with partial medial facetectomy and foraminotomy. Microscope used.

---

🧾 Feedback Workflow (Dr. Pittman)

Every time you use Vertebro, please:

1. Review the CPT output table.
2. Compare it with the true billing sheet or expected codes.
3. Provide feedback such as:
   • ✅ “Correct CPTs and rationale.”
   • ⚠️ “Partially correct — missing code X or modifier Y.”
   • ❌ “Incorrect — CPT Z should not appear.”
4. Optionally add notes on phrasing not recognized (e.g., “MIS lateral approach not detected”).

This feedback will be used to iteratively refine and retrain the rule logic and future ML models.

---

📈 Testing & Exploration Areas

Region	Case Type	Focus
Cervical	ACDF, CDA, Posterior Fusion	Fusion + anterior plating logic
Thoracic	Decompression, Instrumentation	Mid-thoracic handling
Lumbar	TLIF, PLIF, ALIF, Post-lat Fusion	Interbody vs posterolateral overlap
Any	Revision, Hardware Removal	22849, 22852, staged operations
Multi-Level	Complex constructs	Code chaining: 22552, 22614
Graft / Instrumentation	Autograft, Allograft, Plate/Rods	Add-ons: 20936, 20930, 22842

---

📬 Contact

For questions, case feedback, or feature suggestions, please reach out to:
📧 slaiwala@bidmc.harvard.edu
👤 Developer: Shubham Laiwala
🏥 Affiliation: Nazarian Lab, BIDMC / Harvard Medical School

---

🩺 Short Summary (for Hugging Face “About” sidebar)

Vertebro automatically suggests CPT codes from spine operative notes using a rule-based NLP engine.
Designed for research, billing validation, and clinical education.
Feel free to test real, de-identified operative notes!