Clinical Obsidian

• 

  training model

• 

  main page of the app

• 

  check my risk it's here the patient can check he's health issue

• 

  this upload part is for patient to ask for diagnosis it can be ct scan, Xray or any other medical document

• 

  check my risk result

• 

  upload side output for medical record

Inspiration

This project was inspired by how AI can help doctors make better decisions using patient data. In real hospitals, doctors have limited time, so a system that gives quick and reliable predictions can be very useful.

I wanted to build something more than a basic ML model and try to make a system that is closer to real healthcare applications.

To make the system more practical, I added a feature where users can upload medical documents, and the system automatically extracts important health information from them.

Medical Document Upload Feature

Users can upload different types of medical files:

• PDF Reports Lab results, discharge summaries

• CT / X-Ray Images Scan images (JPG, PNG)

• ECG Reports Heart rhythm printouts

• Blood Test Reports Any lab report image

The system processes these documents using AI to extract useful medical data, which can then be used for prediction.

How I Built the Project

I built a Clinical Decision Support System (CDSS) that predicts disease risk (like heart disease or diabetes) based on patient data.

The workflow is:

Input Data / Medical Documents → Data Processing → ML Model → Prediction → Explanation → Output

Models Used

• Logistic Regression
• K-Nearest Neighbors (KNN)
• Random Forest
• Gradient Boosting

Data Processing

• Handled missing values
• Scaled features
• Encoded categorical data
• Extracted structured data from uploaded medical documents

Probability and Risk Levels

Instead of just Yes/No, the system gives probability and risk levels:

• Low Risk: (0 - 0.3)
• Medium Risk: (0.3 - 0.7)
• High Risk: (0.7 - 1.0)

Explainable AI

I used SHAP to explain predictions. It shows which features (like age, BP, glucose) affected the result.

Prediction = f(features)

SHAP helps understand how each feature contributes.

Backend

I used FastAPI to create APIs:

• /predict → gives prediction
• /upload → processes medical documents
• /health → checks system status

Logging

The system stores:

• Input data
• Extracted medical data
• Prediction
• Time

This helps track and review results.

What I Learned

• How to build a complete ML pipeline
• Importance of data preprocessing
• How to extract data from medical documents using AI
• Difference between accuracy and real reliability
• Basics of explainable AI (SHAP)
• How backend APIs work

Challenges Faced

• Handling missing and messy data
• Extracting accurate data from different medical documents
• Choosing the best model
• Understanding model predictions
• Making the system more realistic instead of just a demo

Conclusion

This project helped me understand how AI can be used in healthcare. It is not just about prediction, but also about data extraction, accuracy, explanation, and trust.

Built With

• docker
• fastapi
• git
• joblib
• matplotlib
• numpy
• pandas
• python
• scikit-learn
• seaborn
• shap
• sqlite
• xgboost