Michael Nix

  • Clinical Scientist
  • Leeds Cancer Centre - Radiotherapy

Biography

I am a research focussed clinical scientist, working in advanced imaging and image processing for radiotherapy.

My interests include translating cutting-edge methods into routine clinical practice and maximising patient benefit through digital technology, particularly artificial intelligence. Robustness is critical for novel digital technologies in the health sphere, to enable safe and effective uptake of AI.

My clinical work involves imaging patient motion during radiotherapy. Before changing career into radiotherapy I was a chemical physicist working in academic research. I enjoy the outdoors and live in Leeds with my wife and two boys.

AutoConfidence: Automatic confidence assessment for deep learning in healthcare

Prostate cancer affects 12% of men, predominantly over the age of 50 and treatment options include radiotherapy. In order to deliver radiation safely and effectively, tumours and surrounding organs must be accurately drawn (contoured) on CT scans. This manual process is extremely time-consuming and can be inconsistent.

Artificial intelligence (deep-learning auto-contouring) potentially allows fast, automatic, consistent contouring of tumours and organs. However AI methods are often ‘black-box’ and can result in unpredictable, localised, patient-specific errors and uncertainties. These problems currently make manual checking of AI contours necessary, reducing the benefits of automation.

The ‘AutoConfidence’ project will explore independent AI methods, to automatically identify and quantify contour errors and uncertainties resulting from deep-learning auto contouring. We also intend to predict the impact of these errors and uncertainties on the radiotherapy dose. By doing this, we can identify regions where the dose is significantly affected for human review and intervention, with a traffic light system. This will minimise the burden of manual time needed and maximise the benefits of AI in radiotherapy contouring.

Our approach will also enable automation with robustness, allowing increased clinical confidence in what is otherwise a ‘black-box’ method.

What will help to make digital transformation a reality in the NHS?

Transformative uptake of AI in healthcare requires demonstrable technological robustness, allowing clinical confidence. When coupled with clear patient benefit, workflow efficiency gains and a low activation barrier for training and uptake, this approach will enable and maximise digital transformation potential.