YAML Metadata Warning:The pipeline tag "time-series-classification" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
KaliSense AI — ECG-based Serum Potassium Anomaly Detector
🏆 2025 SelectUSA Investment Summit — Global 2nd Place
🏥 Invited for evaluation by world top-10 medical institutions
🏢 By Precise Bigdata生技 | Precise Intelligent Biotech
Model Description | 模型說明
KaliSense AI is a CNN-Transformer deep learning model that predicts serum potassium (K⁺) anomalies directly from 12-lead ECG waveforms — without any blood draw.
The model classifies ECG inputs into three categories:
- ✅ Normal — K⁺ within normal range (3.5–5.0 mEq/L)
- ⚠️ Hypokalemia — Low potassium risk (K⁺ < 3.5 mEq/L)
- 🚨 Hyperkalemia — High potassium risk (K⁺ > 5.5 mEq/L)
Clinical Background | 臨床背景
Hyperkalemia and hypokalemia are among the most dangerous electrolyte imbalances, especially in:
- Chronic Kidney Disease (CKD) patients
- Hemodialysis patients (9万+ in Taiwan alone)
- Patients on ACE inhibitors, ARBs, or potassium-sparing diuretics
Traditional serum potassium testing requires venous blood draw and lab processing (30 min – hours). KaliSense AI enables real-time, non-invasive potassium risk screening from ECG alone.
Model Architecture | 模型架構
Input: 12-lead ECG (500 Hz sampling rate)
│
▼
CNN Encoder (local feature extraction)
• T-wave amplitude detection
• QRS complex width measurement
• PR interval analysis
• QTc calculation
│
▼
Transformer Encoder (global temporal context)
• Multi-head self-attention (8 heads)
• Positional encoding for temporal sequence
│
▼
Classification Head
• 3-class output: Normal / Hypokalemia / Hyperkalemia
• Softmax probability scores
Performance | 模型效能
Validated on clinical dataset from nephrology and hemodialysis centers (IRB approved):
| Metric | Value |
|---|---|
| Overall Accuracy | 91.2% |
| Hyperkalemia Sensitivity | 88.0% |
| Hyperkalemia Specificity | 93.5% |
| Hypokalemia Sensitivity | 85.4% |
| AUC (Hyperkalemia) | 0.947 |
| AUC (Hypokalemia) | 0.921 |
Input Format | 輸入格式
| Property | Specification |
|---|---|
| ECG Leads | 12-lead (or 1/3/6-lead) |
| Sampling Rate | 500 Hz (recommended) |
| Duration | 10 seconds standard |
| File Formats | .CSV, .EDF, HL7 |
Intended Use | 適用範圍
✅ Intended for:
- Clinical potassium risk screening support
- Hemodialysis patient continuous monitoring
- Rural clinic and remote care settings
- Emergency department rapid assessment
❌ Not intended for:
- Replacing laboratory serum potassium testing
- Standalone diagnosis without physician review
- Pediatric patients (not validated)
Limitations & Disclaimer | 限制與免責聲明
⚠️ This model is intended as a clinical decision support tool only.
All outputs must be confirmed by a licensed physician before clinical action.
Currently under TFDA SaMD regulatory review for medical device software certification.
Privacy & Compliance | 隱私與法規
- Model trained on IRB-approved, de-identified clinical ECG data
- Core model weights and training pipeline stored in private repository under TFDA SaMD compliance controls
- Patient data is never stored beyond the analysis session
- Compliant with ISO 27001 information security standards
Pre-training Datasets | 預訓練資料集
| Dataset | Records | Usage |
|---|---|---|
| PhysioNet PTB-XL | 21,799 ECGs | Pre-training |
| MIMIC-IV-ECG | 800,000+ ECGs | Feature extraction pre-training |
| Internal Clinical (IRB) | 2,847 ECGs | Fine-tuning & validation |
Citation | 引用
If you reference this work, please cite:
@software{kalisense2025,
title = {KaliSense AI: Non-Invasive Serum Potassium Anomaly Detection},
author = {Precise Intelligent Biotech Co., Ltd.},
year = {2025},
url = {https://huggingface.co/precise-biotech/kalisense-ecg-potassium}
}
Contact & Collaboration | 合作洽詢
🏢 Precise Bigdata生技股份有限公司 | Precise Intelligent Biotech Co., Ltd.
📧 contact@precisebiotech.ai
🌐 www.precisebiotech.ai
💻 GitHub: github.com/curry1982/kalisense-ai