How tbachu/pneumoniadetection works
PneumoniaDetection combines dataset acquisition and binary or multi-label model training with image inference, Grad-CAM explanations, and a clinical analysis API/dashboard. The principal user workflow is to upload a chest X-ray, receive model findings and visual explanations, and review clinical context in the dashboard. Training, calibration, and evaluation form supporting model-development stages. The API and clinical engine excerpts are partial, so links are limited to documented behavior and visible calls; unsampled endpoint wiring is not expanded.
Main entry points: app.py FastAPI service · modeling.py Modeling runtime · app.js Dashboard logic · artifacts Model checkpoint
Components 19
Model Development
- Dataset download
data_loader.py - Binary training
train.py - 14-label training
kaggle_multilabel_train.py - Modeling runtime
modeling.py - Model checkpoint
artifacts
Clinical Analysis
- Calibration
calibrate.py - Clinical evaluation
clinical_eval.py - Calibration state
artifacts - Clinical analysis
clinical_engine.py - Clinical report
report_generator.py - Grad-CAM tool
gradcam.py
Prediction Service
- FastAPI service
app.py
Clinician Interface
- Dashboard logic
dashboard/app.js - Dashboard page
dashboard/index.html - Dashboard styling
dashboard/styles.css
Other components
- Clinician
- Kaggle dataset
- Chest X-ray
- Analysis response
Connections 23
- Kaggle dataset → Dataset download: provides dataset
- Dataset download → Binary training: makes data available
- Binary training → Modeling runtime: trains through
- Binary training → Model checkpoint: saves checkpoint
- 14-label training → Model checkpoint: saves checkpoint
- Calibration → Modeling runtime: loads model
- Calibration → Calibration state: writes calibration
- Clinical evaluation → Modeling runtime: loads model
- FastAPI service → Modeling runtime: loads checkpoint
- FastAPI service → Model checkpoint: reads checkpoint
- FastAPI service → Calibration state: loads calibration
- FastAPI service → Clinical analysis: initializes analyzer
- FastAPI service → Clinical analysis: analyzes image
- Clinical analysis → Modeling runtime: preprocesses and explains
- FastAPI service → Clinical report: formats report
- Grad-CAM tool → Modeling runtime: predicts and explains
- Grad-CAM tool → Model checkpoint: loads checkpoint
- Clinician → Dashboard logic: uploads image
- Dashboard page → Dashboard logic: hosts interface
- Dashboard logic → FastAPI service: requests analysis
- FastAPI service → Analysis response: returns findings
- Analysis response → Dashboard logic: renders results
- Chest X-ray → FastAPI service: uploaded for analysis