PhD Candidate:
Mohammad Amin Fathollah Maghsoudi
|
Advanced Nanomaterials for Improved Wound Dressing
|
ARTIFICIAL INTELLIGENCE FOR SMART WOUND CARE
|
A dressing that senses is only half of the solution. Once a wound is reporting its own pH, temperature, moisture and colour, continuously and for days at a time, the limiting factor becomes interpretation: a clinician cannot watch a stream of raw signals, and a colour change means little without a quantitative reading behind it. Equally, the design space of a modern dressing, spanning nanoparticle chemistry, loading, polymer ratio and fabrication parameters, is far too large to explore by trial and error. Machine learning addresses both problems, and it is the thread that connects our materials work to the clinic. |
|
Reading the Wound from an Image
Our dressings carry a colorimetric sensing window whose colour shifts with wound pH, an early and reliable marker of infection. Translating that colour into a number is a computer vision problem: ambient lighting, camera white balance and viewing angle all distort the measurement. We are developing image processing and learned colour-correction models that normalise a smartphone photograph against on-dressing reference patches and return a calibrated pH value, so that a nurse, a carer or the patient can read the wound without removing the dressing. The same imaging pipeline supports automatic segmentation of the wound boundary, measurement of area over time, and classification of tissue type, turning a series of casual photographs into an objective healing trajectory.
Predicting Infection Before It Is Visible
Individual sensor readings are noisy; their combined trend is not. By fusing pH, temperature and moisture as multivariate time series, models can be trained to recognise the signature of bacterial colonisation days before the clinical signs appear, and to distinguish it from the transient inflammation that accompanies normal healing. Our emphasis is on explainable models, which show the clinician which signals drove an alert rather than issuing an unexplained warning; this follows directly from our earlier work using explainable deep learning for tumour detection in robotic surgery.
Designing Materials with Machine Learning
The group has an established record of using artificial intelligence to discover and optimise functional materials. We have applied machine learning to the discovery of ion-selective membranes, to the prediction of enzyme loading, activity retention and reusability in enzyme–metal-organic framework biocatalysts, and to the mechanistic analysis of catalytic CO₂ electroreduction, and we have used soft computing methods to optimise the geometry and response of 3D-printed piezoresistive sensors. We now apply the same approach to wound dressings, using data-driven models to relate composition and process parameters, including nanoparticle loading, polymer ratio and electrospinning conditions, to the properties that matter clinically: porosity, swelling, water vapour transmission rate, degradation rate, antibacterial efficacy and mechanical integrity. The aim is to reduce the number of experimental iterations required to reach a formulation that performs.
Towards Personalised and Closed-Loop Treatment
Ultimately these strands converge on a dressing that acts on what it measures. Because our platform includes stimulus-responsive therapy, in the form of piezoelectric and sonodynamic antibacterial activation and pH-triggered drug release, a model that recognises the onset of infection can also decide when treatment should be delivered and at what intensity. This moves wound care from a fixed dressing-change schedule toward a personalised, closed-loop therapy guided by the wound itself.
Our dressings carry a colorimetric sensing window whose colour shifts with wound pH, an early and reliable marker of infection. Translating that colour into a number is a computer vision problem: ambient lighting, camera white balance and viewing angle all distort the measurement. We are developing image processing and learned colour-correction models that normalise a smartphone photograph against on-dressing reference patches and return a calibrated pH value, so that a nurse, a carer or the patient can read the wound without removing the dressing. The same imaging pipeline supports automatic segmentation of the wound boundary, measurement of area over time, and classification of tissue type, turning a series of casual photographs into an objective healing trajectory.
Predicting Infection Before It Is Visible
Individual sensor readings are noisy; their combined trend is not. By fusing pH, temperature and moisture as multivariate time series, models can be trained to recognise the signature of bacterial colonisation days before the clinical signs appear, and to distinguish it from the transient inflammation that accompanies normal healing. Our emphasis is on explainable models, which show the clinician which signals drove an alert rather than issuing an unexplained warning; this follows directly from our earlier work using explainable deep learning for tumour detection in robotic surgery.
Designing Materials with Machine Learning
The group has an established record of using artificial intelligence to discover and optimise functional materials. We have applied machine learning to the discovery of ion-selective membranes, to the prediction of enzyme loading, activity retention and reusability in enzyme–metal-organic framework biocatalysts, and to the mechanistic analysis of catalytic CO₂ electroreduction, and we have used soft computing methods to optimise the geometry and response of 3D-printed piezoresistive sensors. We now apply the same approach to wound dressings, using data-driven models to relate composition and process parameters, including nanoparticle loading, polymer ratio and electrospinning conditions, to the properties that matter clinically: porosity, swelling, water vapour transmission rate, degradation rate, antibacterial efficacy and mechanical integrity. The aim is to reduce the number of experimental iterations required to reach a formulation that performs.
Towards Personalised and Closed-Loop Treatment
Ultimately these strands converge on a dressing that acts on what it measures. Because our platform includes stimulus-responsive therapy, in the form of piezoelectric and sonodynamic antibacterial activation and pH-triggered drug release, a model that recognises the onset of infection can also decide when treatment should be delivered and at what intensity. This moves wound care from a fixed dressing-change schedule toward a personalised, closed-loop therapy guided by the wound itself.
|
|