TB Detect:
- Convolution Neural Network is used to detect TB in patients using X-ray. The user needs to upload their chest X-ray and the system gives them probability of them having TB.
- The system was trained and tested on two publicly available datasets: Sbenzhen chest X-ray set and Montgomery Country chest X-ray set (MC). Accuracy of 85 percent was achieved.
- The patient can use this web application as a second opinion to confirm diagnosis.
Code and Resources Used
Python Version: 3.7
Packages: pandas, numpy, matplotlib, sklearn, tensorflow , flask, json, pickle
For Web Framework Requirements: pip install -r requirements.txt
Data Set
Dataset is downloaded from kaggle.com The data in in data.csv. The number of columns is 30 and the number of rows is 570. The columns are: For more detail info please visit :https://lhncbc.nlm.nih.gov/publication/pub9931
Model Building
First, I have scaled the data with MinMaxScaler. I also split the data into train and tests sets with a test size of 20%.
I have applied Convolution Neural Network (CNN) with two Convolution layers:
- Output Layer – Units=1, activation=binary crossentropy
I have also applied early stopping.
Model performance
- Validation loss : 0.0115
- Validation Accuracy : 0.85 or 85%
Productionization
In this step, I built a flask API endpoint that is hosted on Heroku. The API endpoint takes in a request with a X-Ray scan of lungs and returns whether the TB is detected or not.
Application link: https://tb-detector.herokuapp.com/
