Artificial Intelligence Based Lung Sound Classification

Project Title

Artificial Intelligence Based Lung Sound Classification

Project Area of Specialization Biomedical EngineeringProject Summary

Respiratory diseases are a leading cause of death in the world and accurate lung auscultation is extremely important for the diagnosis and evaluation of patients. However, this method is vulnerable to physician and instrument limitations and there is strong interest in automation of lung sound analysis system.

We are proposing a robust Lung Sound Classification System based on machine learning. Lung Sound Classification Stethoscope will be capable of  analyzing patients. Data will be collected from the patient by the Sensing System of Stethoscope for classification. The recorded data will then be fed to the Amplifier and Filtering ,Feature extraction and Central System (Classification System), where the machine learning model will classify the signal either as normal or abnormal (Binary Classification). Once the result is obtained, this will be displayed on the stethoscope display screen.

Project Objectives

Lungs being a vital organ of the human body, is of great importance and with the outbreak of COVID-19 pandemic, the treatment of lung diseases, which are in the top 10 causes of death are of great importance in today's medical field.

The physician or doctor uses the Conventional Stethoscope (CS) as the traditional method for the auscultation of lungs, to diagnose the disease and to evaluate the patients. Several factors determine the quality of the detection of the lung sounds. Some of them are the patient’s body size, obesity, chest hair, noise level in the room, proper placement of CS on the chest, hearing capacity of the physician and quality of the sounds generated inside the chest. These factors are applicable for both normal and abnormal chest sounds. This method requires high expertise and a well trained physician, if it is not so, this may lead to false diagnosis and could lead to serious condition.

In such a situation, a physician cannot rely only on his conventional stethoscope to make the final decision. Furthermore, the signal hearing by a physician cannot be recorded and interpretation of sound fully depends upon his expertise. In order to overcome such difficulties, an attempt will be made to prototype an electronic stethoscope with several added advantages, like the sounds could be amplified and recorded for further analysis. Moreover such amplified and analyzed sounds will be used for binary classification.

The objective of our project is:

1. To develop a system that can store lungs sounds, collected by the Sensing System.

2. Pre-Processes (Signal Enhancement/De-noising) on the collected data.

3. Classify the pre-processed data based on machine learning models.

4. Display the result.

5. Evaluate the patients automatically based on the classification.

Project Implementation Method

The proposed system for LSC to classify and diagnose the patients lung sound will be based on the following sections:

1. The Sensing System:This Section will be responsible for recording of lung signals (LS). This section consists of a transducer such as piezoelectric ceramic plates, used for recording sounds.
2. Amplifier and Filters:Amplification and filtering are the two major aspects in any signal acquisition system. Usually, a pre-amplifier with a small gain will be used to suppress the 50 or 60 Hz interference from power lines. An anti-aliasing filter will be used to prevent aliasing effects. The filter section will be built with a band pass filter having the frequency range of most LS signals. In post amplification, the filtered signal will be amplified to the level range required by the ADC.
3. Analog-to-Digital converter (ADC):The amplified and filtered analog signal will be converted to digital signal by ADC. The rate and bit resolution can be set to at least twice the highest frequency of Interest (5 kHz in the case of lungs) in order to preserve sound quality during recording.
4. Pre-Processing:

   De-Noising:A digital filter is sometimes used to extract the signal within the frequency band of interest from the noisy data. In order to equip the system with even better denoising capability, some advanced artifacts removal techniques will be utilized such that the output signal-to-noise ratio (SNR) can be further improved.

   Normalization and Segmentation: In data acquisition, different sampling and acquisition locations normally result in a signal variation. Thus, the LS signals will be normalized to a certain scale, so that the expected amplitude of the signal does not affect from, the data acquisition locations. After getting the normalized signals, the LS signals will be segmented into cycles which will be ready for LS features extraction and classification.
5. Central System (Signal Processing):

   Feature extraction:Signal processing will be carried out to convert the raw data to some type of parametric representation, called feature, will be then used for further analysis, processing and finally for classification.

   Classification:A classifier, trained with the extracted features, will be used to categorize the data and assist the medical specialist for the diagnosing and evaluation of patients.

   The Central System will have a Machine Learning (ML) model running to classify whether the signal has any abnormalities, based on the trained data. The proposed model will be trained on the existing publicly available data of lungs recordings to do binary classification. it will go through a testing phase using local real-time testing data.

6. Result:The output of the ML model will decide the result of the lung recording. Once any abnormality is detected the model will then evaluate the patients.

7. Display:Result of the system will be displayed on the LCD screen.

Benefits of the Project

This project will provide a unique and easy-to-use system for classification of lung sounds and will serve as an assisting tool for pulmonologists all over Pakistan. This system will enhance the overall testing and diagnosing process of the lung patients. The designed biomedical system will be in high demand as the pulmonological diseases are the major causes of deaths in Pakistan and over the world, infact many medical institutes will be looking to implement this system in their institutes for accurate purposes.

Technical Details of Final Deliverable

The final deliverables of this project will be the following sub modules:

1. Sensing, Amplification and Filtering, ADC and Pre-Processing System

2. Center System

3. Machine Learning Algorithms

Sensing, Amplification and Filtering, ADC and Pre-Processing System:

Sensing system contains a piezoelectric ceramic plate, amplifier and filter to amplify and to filter out the noise and Anolog-to-Digital Converter.

Center System:

The lung sound recordings from the above system will then be processed in the Center System, here it will have a machine learning model running to classify lung signals into normal or abnormal ones. Once classified, the results will be displayed on the monitoring screen.

Machine Learning Algorithms:

Neural network (NN) finds a role in a variety of applications due to the combined effect of feature extraction and classification availability in deep learning algorithms. Here convolutional Neural Network (CNN) and Recursive Neural Network (RNN) will be utilized to make the  system smart enough to classify different lung sound signals based on the input signals. 

CNN signal classification model takes an input signal, processes it and classifies it under certain categories (Eg., Normal, Abnormal). Computers see an input signal as an array of sampling points and it depends on the signal resolution.

The recurrent structure of RNN makes it capable of learning and making full use of the temporal information of the input signals to make up for the deficiencies of the short-term features

Deep learning CNN and RNN models will train and test each input signal and pass it through a series of convolution layers with filters, Pooling, fully connected layers (FC) and apply different functions to classify the given signals.

Final Deliverable of the Project HW/SW integrated systemCore Industry HealthOther Industries IT , Medical Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Good Health and Well-Being for People, Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
Sensing System	Equipment	1	10000	10000
Hardware System	Equipment	1	20000	20000
Amplifying System	Equipment	2	5000	10000
ADC	Equipment	1	7000	7000
GPU	Equipment	1	10000	10000
LCD Display	Equipment	2	5000	10000
Batteries(rechargeable)	Equipment	2	1500	3000
Surveying, Traveling etc	Miscellaneous	1	10000	10000