Datheory/medsync-indobert-inacbg

🏥 MedSync AI Studio: Medical-IndoBERT Dual-Head INA-CBGs Sync

Model klasifikasi teks klinis multi-task berbasis Medical-IndoBERT (Domain-Adaptive Pretrained) yang disempurnakan melalui Supervised Fine-Tuning (SFT) dengan arsitektur Dual-Head:

1. Head 1 (ICD-10): Klasifikasi Diagnosis Klinis Primer (WHO ICD-10).
2. Head 2 (INA-CBGs): Prediksi Kelompok Tarif Casemix Klaim (Kemenkes RI / UNU-IIGH Casemix).

---

📋 Detail Arsitektur & Pelatihan

• Backbone: indobenchmark/indobert-base-p2 + Domain-Adaptive Pretraining (DAPT) pada korpus medis klinis.
• Arsitektur: DualHeadModel (Encoder BERT + Dropout 0.2 + 2 Linear Classification Heads).
• Jumlah Kelas ICD-10: 45 kelas diagnosis.
• Jumlah Kelas INA-CBGs: 46 kelompok tarif casemix.
• Studi Kasus: RSUD Datu Sanggul (Kelas C, Regional 4).
• Kepatuhan Privasi: UU Perlindungan Data Pribadi (UU PDP) No. 27 Tahun 2022.

---

🚀 Cara Penggunaan Singkat (Python / Google Colab / Notebook)

Senior atau peneliti dapat langsung menguji model ini menggunakan kode berikut:

import torch
import torch.nn as nn
from transformers import AutoConfig, AutoTokenizer
from huggingface_hub import hf_hub_download
import json

REPO_ID = "Dragneva/medsync-indobert-inacbg"

# 1. Download Model Weights & Metadata dari Hugging Face Hub
model_path = hf_hub_download(repo_id=REPO_ID, filename="model.pt")
mapping_path = hf_hub_download(repo_id=REPO_ID, filename="label_mappings.json")

with open(mapping_path, "r", encoding="utf-8") as f:
    meta = json.load(f)
id_to_icd10 = {int(k): v for k, v in meta["id_to_icd10"].items()}
id_to_group = {int(k): v for k, v in meta["id_to_group"].items()}
icd10_to_name = meta.get("icd10_to_name", {})
group_to_name = meta.get("group_to_name_id", {})

# 2. Inisialisasi Arsitektur DualHeadModel
class DualHeadModel(nn.Module):
    def __init__(self, config, n_icd, n_grp):
        super().__init__()
        from transformers import AutoModel
        self.encoder = AutoModel.from_config(config)
        self.dropout = nn.Dropout(0.2)
        self.icd10_head = nn.Linear(config.hidden_size, n_icd)
        self.group_head = nn.Linear(config.hidden_size, n_grp)

    def forward(self, input_ids, attention_mask):
        outputs = self.encoder(input_ids=input_ids, attention_mask=attention_mask)
        mask = attention_mask.unsqueeze(-1).float()
        pooled = (outputs.last_hidden_state * mask).sum(dim=1) / mask.sum(dim=1).clamp(min=1e-9)
        return self.icd10_head(self.dropout(pooled)), self.group_head(self.dropout(pooled))

# 3. Load Model & Tokenizer
config = AutoConfig.from_pretrained(REPO_ID)
tokenizer = AutoTokenizer.from_pretrained(REPO_ID)
model = DualHeadModel(config, n_icd=len(id_to_icd10), n_grp=len(id_to_group))
model.load_state_dict(torch.load(model_path, map_location="cpu"))
model.eval()

# 4. Inferensi Teks Klinis
teks_klinis = "Pasien datang kontrol rutin poli penyakit dalam dengan riwayat hipertensi 5 tahun. Tensi 150/90 mmHg."
inputs = tokenizer(teks_klinis, return_tensors="pt", max_length=128, truncation=True, padding=True)

with torch.no_grad():
    icd_logits, grp_logits = model(inputs["input_ids"], inputs["attention_mask"])
    icd_probs = torch.softmax(icd_logits, dim=-1)[0]
    grp_probs = torch.softmax(grp_logits, dim=-1)[0]

top_icd = torch.argmax(icd_probs).item()
top_grp = torch.argmax(grp_probs).item()

icd_code = id_to_icd10[top_icd]
grp_code = id_to_group[top_grp]

print("=== HASIL INFERENSI MEDSYNC AI STUDIO ===")
print(f"Prediksi ICD-10 : {icd_code} - {icd10_to_name.get(icd_code, '')} ({icd_probs[top_icd]*100:.1f}%)")
print(f"Prediksi INA-CBGs: {grp_code} - {group_to_name.get(grp_code, '')} ({grp_probs[top_grp]*100:.1f}%)")

---

🔒 Etika & Kepatuhan Data

Dataset dan bobot dikembangkan sesuai regulasi perlindungan data medis UU PDP No. 27/2022. Rekam Medis (Nomor Rekam Medis, No. SEP, Nama Pasien) telah dideidentifikasi dan disimpan di repositori privat terpisah.