Interictal epileptiform discharges (IEDs) in electroencephalograms (EEGs) are a biomarker of seizure detection in epilepsy and are useful in supporting the diagnosis of epilepsy. However, due to the extensive types of morphologies of IEDs and their similarity to waves that are part of artifacts, the detection of IEDs is not easy. Moreover, there is a dearth of qualified experts to interpret EEG results. Various studies have been done in Pakistan to automate the detection of IEDs, but they have been limited by small samples and thus have not demonstrated expert-level performance. Therefore, there is a need for a validated automated method to detect IEDs with expert-level reliability.

The objectives of this study are to assess if the algorithm analyses EEG with accuracy comparable to that of physicians with a specialty in clinical neurology at Shifa International Hospital (SIH), thereby decreasing the chances of human error.

This study will proceed in 3 phases. First, we will identify the best algorithms available in open source. Install these algorithms and understand their performance on data available in open source. From these, we will select and implement 2 or 3 best algorithms for pre-processing, feature extraction and IED detection. Second, we will acquire data from Shifa Neurology Lab which will have the annotated IEDs.Third, evaluate the performance of selected algorithms on data acquired prospectively from Shifa Neurology lab, thereby enhancing these algorithms.

This project will develop solution to assist physicians detect pneumonia accurately. Mobile phones with built in as well as attached sensors will be used to measure and record patient’s symptoms. State of the art machine learning classifiers will be programmed to run on the mobile phone and assist physicians in accurate pneumonia diagnosis. In order to build these classifiers, patient data will be collected and annotated by physicians under a well-defined and approved clinical experiment protocol. Developed system will also have state of the art supervised learning capabilities. For each recommendation given by the system, physicians’ diagnosis will also be recorded and classifiers will continually learn.

Following patient symptoms will be collected using mobile device built in sensors and attached sensors where available:

1. Lung sound
2. Cough sound
3. iii. Temperature
4. Blood oxygen saturation level
5. Patient history
6. Chest X-Ray (if ordered by the physician)
7. vii. Blood tests (if ordered by the physician)

Classifiers will be designed to be effective in absence of symptom measurements and tests results that are not available. As described in next section, classifiers built using annotated cough samples and lung sounds can give rather accurate diagnosis of childhood pneumonia. An attempt will be made to synthesize symptom measurement using Bayes’ nets.

Final solution will be made available to healthcare system throughout the country.

State of the art work to measure cardiac function is described in paper Video-based AI for beat-to-beat assessment of cardiac function by David Ouyang, Bryan He, Amirata Ghorbani, Neal Yuan, Joseph Ebinger, Curt P. Langlotz, Paul A. Heidenreich, Robert A. Harrington, David H. Liang, Euan A. Ashley, and James Y. Zou. Nature, March 25, 2020. https://doi.org/10.1038/s41586-020-2145-8

Source code for application titled “EchoNet-Dynamic” and data to train deep learning neural networks are available at https://github.com/echonet/dynamic

EchoNet-Dynamic is an end-to-end beat-to-beat deep learning model for

1. semantic segmentation of the left ventricle
2. prediction of ejection fraction by entire video or subsampled clips, and
3. assessment of cardiomyopathy with reduced ejection fraction.

This project will evaluate performance of these models on echocardiographs collected at Shifa Cardiology Lab.  Models will be optimized for local data.

In Mitral valve prolapse and regurgitation, the flaps (leaflets) of the mitral valve don't close tightly, causing blood to leak backward into the left atrium of your heart. If not treated, it can result in heart muscle damage. This condition is commonly caused by mitral valve prolapse, in which the leaflets bulge back into the left atrium as your heart contracts.

In Mitral valve stenosis, the flaps of the mitral valve become thick or stiff, and they may fuse together. This results in a narrowed valve opening and reduced blood flow from the left atrium to the left ventricle.

Tracking the mitral valve leaflet in an ultrasound sequence is a challenging task because of the poor image quality and fast and irregular leaflet motion. Usually algorithms apply standard segmentation methods based on edges, object intensity and anatomical information to segment the mitral leaflet in static frames. However, they are limited in practical applications due to the requirement of manual input for initialization or large annotated datasets for training.

EchoNet-Dynamic application available at https://github.com/echonet/dynamic, gives accurate segmentation of left ventricle.

It is also known that the image sequence of a cardiac cycle can be well approximated with a low-rank matrix, except for the mitral leaflet region with fast motion and tissue deformation. Based on this difference, and accurate left ventricle segmentation we propose to track the mitral leaflet by detecting contiguous outliers in the low-rank representation.