Bohndiek Group

Enabling optical interrogation of complex biological environments

Making measurements at a range of depths in a living breathing patient in a noisy, bright, busy clinical environment presents a profoundly different set of constraints to those commonly found in optics laboratories. For hyperspectral imaging this presents a major challenge, since any distortions or artifacts in the recorded 3D ‘hypercube’ (x,y,wavelength) will lead to misclassification of tissue types.  

 We have spearheaded efforts to address several of the main factors leading to such distortions, namely motion, miniaturization, and calibration, through advances in physics, chemistry, and data science. By introducing new high-speed line-scan and custom snapshot hyperspectral systems while applying deep-learning analyses we were able to build distortion-free hypercubes – even for handheld operation – and perform real-time classification. We have conceived new optical filter fabrication methods and endoscope designs to miniaturize hyperspectral imaging systems and enable imaging through hair-thin optical fibers.  

Importantly for clinical translation, we have led international efforts in standardisation: establishing criteria for biophotonic standards; developing new materials for stable and dynamic tissue-mimicking test objects; and leading global multi-centre evaluations. 

Deriving meaning from imaging measurements to observe biological phenomena

In a clinical context, the interpretation of an image leads to decision-making that impacts patient management. The simplest and easiest optical interaction to measure in humans is the absorption of light by the hugely abundant molecule haemoglobin, which gives red blood cells their colour and allows us to interrogate vascular architecture and function using optics. Equipped with new imaging tools, we are creating a deeper understanding of vascular alterations associated with early pre-cancerous changes and how tumours create and modify their vascular structures to derive oxygen and nutrients during early cancer and in response to therapy. Most recently, we have begun to parameterise these vascular measurements, using both biophysical modelling and machine learning methods, to predict how changes in vessel size and topology contribute to the observable macroscopic effects of dynamic hypoxia exposure. 

Cumulatively, our studies, in both cancer models and patients, have caught the attention of the academic community and the public, receiving widespread international media attention.