Classification of data read form MEMS cancer (Impedance) sensor by a Multi-Layer-Perceprton (MLP) Classifier
This work studies the impedance verses frequency characteristics of a blood cell to detect the level of malignancy in it. The developed MEMS sensor is connected to Analog Device's AD5933 impedance analyzer, the AD5933 after performing a frequency sweep from 1kHz to 100kHz reports the impedance data (100 samples) back to an Arduino over the i2c bus. The impedance data so collected is then passed on to a MLP classifier to print out the degree of malignancy in a given sample of human blood. The MPL classifier used here is two layes deep and uses 100 input neurons. The MLP classifier is implemented on Python 2.7 using sklearn on a Raspberry Pi zero W running Raspbian Lite.
To get started copy the ann_ad5933_serial_support.py, sensor_data_cancer_test.csv and sensor_data_cancer_train.csv files to a desired folder. The Arduino must be kept conneted to the h/w UART of the Raspberry Pi zero W via a TTL level shifter (as the Pi and Arduino uses different TTL levels).
The support modules required to run the code are :
1. NumPy --> For numerial calculations
2. SciPy --> For scientific calculations
3. Pandas --> To prase the CSV files
4. Sklearn --> To implement the neural network
To install all the above dependencies have the latest version of "pip => 1.5.4" installed. (check: $pip -V):
1. To install NumPy --> sudo pip install numpy
2. To install SciPy --> sudo pip install scipy
3. To install Pandas --> sudo pip install pandas
4. To install Sklearn --> sudo pip install sklearn
After uploading the ad5833_arduino_code to the Arduino, run the code $python ann_ad5933_serial_support.py with the Arduino connectedto the UART of the Raspberry Pi zero W. To check the serial communication port allocated to the Arduino type
$sudo dmesg | grep tty, replace what ever tty value is printed out with the one in the code. After running the code, the program will read and store the impedance data obtained in sensor_data_cancer_test.csv.
After training on the data from sensor_data_cancer_train.csv the prediction is done, the code returns the malignancy state (i.e. Cancerous or Normal) along with the predicted label (i.e. [1] or [0]). To further tune the network change the number of neurons in the hidden layers at clf = MLPClassifier(solver='lbfgs', alpha=1e-5, hidden_layer_sizes=(50, 2), random_state=1)
The graph between THP and PHA cell line impedance is,
The device while displaying the nature of malignancy, Red for Cancerous and Green for normal blood sample ,
- Scikitlearn - The Neural Network by sklearn
- Pandas - Python data management lib
- Debjyoti Chowdhury - Initial work - MyGithub
This project is licensed under the MIT License - see the LICENSE.md file for details