Abstract: 

EEG signal analysis has been of interest for diagnosing epilepsy in hospitals, monitoring treatment efficiency during follow-up visits, and providing continuous care to ambulatory patients. Diagnosis entails seizure event identification in the EEG recordings from the patient. Initial diagnosis, extended diagnosis and follow up procedures leads to storage of large amounts of epileptic patient’s EEG records based on data compression and later retrieval of pertinent data. Patient-care encompasses need to predict and detect seizures at the patient-end in the EEG captured by a wearble sensor to warn the patient and the care provider. These processing tasks are usually realized separately though some of these may be closely related from the utility point of view. This thesis is about realizing two related processing tasks to deliver not only the best and improved performances individually but in which their processing or workflows are significantly synergized or shared. This synergized realization of two related processing tasks also brings with it the processing and resource optimization (through sharing) and may also augment performance efficacy of the individual tasks. This target makes the problem challenging. The thesis demonstrate its methodology in two scenarios: In the context of ambulatory patientcare these are the tasks of seizure detection and prediction. In the context of neurologists support, the tasks considered are seizure detection (automatic annotation) along with compression. The proposed shared framework for EEG detection and prediction is channel-scalable looking to requirements of a wearble-device based patient specific system. This can process a pre-selected single channel or can be built up into a multi-channel system for performance enhancement based on ensembling the results of pre-selected individual channels. The research demonstrates that a hybrid approach comprising of a shallow autoencoder and a conventional classifier can serve as a high performance common workflow for seizure detection and prediction. Resource sharing is indeed the additional dividend. It attains 98.8% accuracy and 99.2% sensitivity for seizure detection, and 99% accuracy and 99.3% sensitivity for seizure prediction using single EEG channel of the CHB-MIT dataset. Multi-channel framework increases the performance to 99.8% accuracy and 99.7% sensitivity for seizure detection, and  99% accuracy and 99.3% sensitivity for seizure prediction. This also exceeds the best results reported in literature at the additional benefit of lowest computational complexity. The thesis advances research on utilizing autoencoders to generate compact trainable latent representation using shallow architecture. Treating the code as a feature set, it is shown that it can be further reduced for effective classification by employing feature reduction techniques. The second direction of this work has resulted in an extended intelligent Neurologists Support System (i-NSS), based on Discrete Wavelet Transform coefficients used both for classification and compression based on traditional techniques. This synergy not only results in data reduction/summarization and adaptive compression as per importance of data but also allows maintaining the classification fidelity of reconstructed signal greater than 99% with the classification results obtained on the original data. In one of our work, we have further extended i-NSS to distinguish between four types of epilepsies. With its ability to scale across different EEG channel configurations, high performance and low computational cost, this work lays the foundation for developing smart, trainable EEG systems for customized, real-time epilepsy management.

Final Thesis Defense Committee:   

• Dr. Muhammad Farhat Kaleem, Associate Professor, Dean School of Engineering, UMT (External Examiner)
• Dr. Mian Muhammad Awais, Professor, CS, LUMS (Member outside the department)
• Dr. Nadeem Ahmad Khan, Associate Professor, EE, LUMS (Supervisor)
• Dr. Muhammad Awais Bin Altaf, Senior Staff ASIC Design Engineer, Microvision, Redmond, Washington (Co-Supervisor)
• Dr. Ijaz Naqvi, Associate Professor, EE, LUMS (Ph.D. Committee member)
• Dr. Hassan Mohy Ud Din, Assistant Professor, EE, LUMS (Ph.D. Committee member)

List of Publications: 

Journals:

1. GH. Khan, NA. Khan, MAB. Altaf, Q. Abbasi, “A shallow autoencoder framework for epileptic seizure detection in EEG signals”, Sensors, vol. 23(8), p.4112, 2023.

2. GH. Khan, NA. Khan, W. Saadeh, MAB. Altaf, “Epileptic Seizure Prediction Based on the Encoded Representation of EEG Signal Epochs by a Shallow Autoencoder”, IEEE Open Journal of Engineering in Medicine and Biology (OJEMB), (submitted)

Conferences

1. GH. Khan, NA. Khan, MAB. Altaf, MUR. Abid, “Classifying Single Channel Epileptic EEG data based on Sparse Representation using Shallow Autoencoder”, 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 2021, pp. 643-646.

2. GH. Khan, NA. Khan, MAB. Altaf, “Shallow sparse autoencoder based epileptic seizure prediction.” In 2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2022, pp. 2349-2356.

3. GH. Khan, NA. Khan, W. Saadeh, MAB. Altaf, “Using Sparse Representation of EEG Signal from a Shallow Sparse Autoencoder for Epileptic Seizure Prediction”, BIOSIGNALS, 2023, pp. 125-132.

4. S. Shakeel, N. Afzal, GH. Khan, NA. Khan, MAB. Altaf, “EDM: A multiclassification support system to identify seizure type using K Nearest Neighbor”, 28th IEEE International Conference on Electronics, Circuits, and Systems (ICECS), 2021, pp. 1-4.

5. GH. Khan, MA. Hashmi, MM. Awais, NA. Khan, RB. Ahmad, “High Performance Multi-class Motor Imagery EEG Classification”, Biosignals, 2020, pp. 149-155.

Book Chapters

1. GH. Khan, NA. Khan, W. Saadeh, MAB. Altaf, “Epileptic Seizure Detection and Prediction for Patient Support”, In International Joint Conference on Biomedical Engineering Systems and Technologies, pp. 40-59. Cham: Springer Nature Switzerland, 2023.

2. NA. Khan, GH. Khan, MA. Ahmad, MAB. Altaf, M. O. Tarar, “The Extended i-NSS: An Intelligent EEG Tool for Diagnosing and Managing Epilepsy”, In International Joint Conference on Biomedical Engineering Systems and Technologies, pp. 243-262. Cham: Springer International Publishing, 2020.