Dr James P Howard

Links

My CV

Google scholar

  • Publications: 131
  • Citations: 1870
  • H-index: 18

Twitter

GitHub

CardiologyTrials.org

About me

I'm a Wellcome Trust PhD Fellow and cardiology registrar at Imperial College London. My thesis is entitled "Machine Learning in Cardiology". Some of my source code is available here on my GitHub and my publications are visible on my Google Scholar profile and CV.

My current work largely focuses on the use of convolutional neural networks in cardiac imaging, and I am a proficient user of the Python programming language and the Pytorch framework. I also have a good understanding of medical statistics, and am occasionally forced to used R.

Highlighted work

Pacemakers

We've developed a free-to-use tool utilising a convolutional neural network which is able to accurately identify the model of pacemaker present on a chest X-ray. This work has been published in JACC: Clinical Electrophysiology (open access).

Arterial waveforms

The identification of adverse pressure signals during an angiogram, termed "damping", is necessary to maintain safety and the accuracy of coronary physiology measurements. We developed a convolutional neural network which is able to accurately monitor these signals. This work has been published in JACC: Cardiovascular Interventions

Cardiac MRI

We investigated whether deep learning of the sequences acquired in the first minutes of a scan could provide an early alert of abnormal features. It does this by segmenting out slices of the heart acquired at the start of a scan, and using these to reconstruct an accurate 3D model of the patient's heart.

Echocardiography

Automated echocardiographic interpretation hinges on the correct recognition of the view (imaging plane and orientation). Current state-of-the-art methods for identifying the view computationally involve 2-dimensional convolutional neural networks (CNNs) and ignore information describing the movement of structures throughout the cardiac cycle. Here we explore the efficacy of novel CNN architectures and find that these can more than halve the error rate of traditional CNNs. The work is published in JMAI.