365MC multi-output ML

A machine learning-based clinical decision support system (CDSS) for predicting postoperative liposuction outcomes.

This repository contains a multi-output prediction pipeline and a Streamlit-based CDSS prototype for individualized postoperative outcome prediction using demographic, surgical, and body composition variables.

---

🎇 Overview

The final deployed model is a chained Extra Trees regressor designed for multi-output prediction of postoperative liposuction outcomes.

The current prototype predicts:

• Postoperative weight
• Postoperative body size

The repository includes:

• trained model artifacts
• model development notebooks
• model interpretation outputs
• a Streamlit-based user interface for prediction

---

📊 Representative Model Performance

The representative final model was a chained Extra Trees regressor, which showed strong predictive performance for postoperative outcome prediction.

Model	Prediction Task	R²	RMSE	MAE	MAPE
Chained Extra Trees	Postoperative outcome prediction	0.980	2.356	1.242	2.199

---

Model Architecture

The final model uses a regressor chain structure with sequential prediction:

1. Extra Tree Regressor 1 predicts postoperative weight
2. Extra Tree Regressor 2 predicts postoperative size using the original input features together with the predicted postoperative weight

This chained structure allows the model to account for the dependency between postoperative outcomes.

---

Input Data

The model was developed using a de-identified multicenter liposuction cohort from the 365MC network in South Korea.

Item	Description
Source	365MC multicenter liposuction registry
Study period	2024
Clinical sites	20 obesity specialty clinics
Final cohort	7,804 eligible individuals
Modeling task	Multi-output regression
Input type	Preoperative demographic, surgical, anthropometric, and body composition variables
Output targets	Postoperative body weight and postoperative circumferential size

---

Input Features

The final model uses 15 preoperative predictors.

Category	Input variables
Demographics	sex, age
Anthropometrics	height, preoperative body weight, BMI, preoperative circumferential size
Surgical information	liposuction technique, liposuction site
Body composition	skeletal muscle mass, body fat mass, total body water, fat-free mass, body protein, body mineral content, waist-to-hip ratio

These variables are entered through the Streamlit CDSS interface and are processed using the saved preprocessing objects before model inference.

---

CDSS Prototype

The Streamlit-based CDSS prototype provides a step-by-step workflow for:

1. Demographics input
2. Liposuction information input
3. Body composition input
4. Prediction output display

The application is designed as a research-oriented prototype for intuitive postoperative outcome prediction.

---

💻 Repo Layout

365MC-multioutput-ml/
|-- app/                 Streamlit CDSS application
|   `-- page.py
|
|-- assets/              UI images
|
|-- data/                De-identified datasets
|   |-- processed/
|   |-- train_x.csv
|   |-- train_y.csv
|   |-- test_x.csv
|   `-- test_y.csv
|
|-- models/              
|   |-- chained_et_final.pkl
|   `-- scaler_bundle.pkl
|
|-- notebooks/           
|   |-- 02-eda.ipynb
|   |-- 03-modeling.ipynb
|   `-- 04-modeling-final-chain-et.ipynb     Main final model notebook       
|
|-- reports/            
|   |-- model_architecture.png
|   |-- cdss_overview.png
|   |-- correlation_matrix.png
|   |-- feature_selection(L1).png
|   |-- shap_et_weight_step1.png
|   `-- shap_et_size_step2.png
|
|-- requirements.txt
`-- README.md

---

Quick Start

Installation

git clone https://github.com/syselina/365MC-multioutput-ml.git
cd 365MC-multioutput-ml
pip install -r requirements.txt

Run the App

streamlit run app/page.py

---

Notes

This repository is intended for research, demonstration, and portfolio purposes.

The included CDSS is a prototype implementation and should be interpreted as a decision-support tool rather than a standalone clinical decision-making system.

---

📝 Manuscript Status & Project Timeline

The associated manuscript is currently under revision at a confidential SCI-indexed journal.
The journal name and detailed submission information are not disclosed in this public repository.

Date	Milestone	Status
2025-04-15	Project initiated	Complete
2025-06-25	Main model development and code implementation completed	Complete
2025-06-25	Manuscript drafting initiated	Complete
2025-08-17	Manuscript submitted to SCI-indexed journal	Complete
2026-05-12	1st revision initiated	Complete
2026-06-01	1st revision completed	Complete
2026-06-30	2nd minor revision initiated	In progress

Current manuscript status: under SCI journal review, second revision in progress.

---

🧑‍💻 Team & Contributions

Area	Main	Assist	Description
Drafting	Selina	Chaewoo	Manuscript and project framing, documentation, and revision
Model development	Selina	Chaewoo	Multi-output ML modeling workflow and final chained Extra Trees model development
CDSS development	Chaewoo	Selina	Streamlit-based CDSS

---

Disclaimer

This application is a research prototype for postoperative liposuction outcome prediction.
It is not intended to replace professional clinical judgment, physician decision-making, or institutional protocols.