Description
Optical microscopy methods are commonly employed to measure rheological properties, like viscosity or viscoelasticity, of materials from biological tissues to colloidal suspensions. Such rheological properties are important for applications such as formulating pharmaceutical ointments to tailoring the mouthfeel of foods and for answering many fundamental questions such as how phase separation proceeds and how the material properties of the cytoplasm respond to stimuli. In recent years, an optical microscopy technique for performing rheological measurements known as differential dynamic microscopy (DDM) has grown in popularity. However, a drawback of DDM is that one typically needs to acquire movies of ~1000 frames of a sample to quantify that sample's properties. If multiple samples or multiple locations within a sample need studying, this is a time-consuming process. Therefore, we devised a method to record a movie of our sample as we scan the sample using a motorized stage. This allows us to measure properties across a large region of the sample (or possibly multiple samples) quickly. However, because the sample is moving as we record images, the computational image analysis becomes more complicated. We discuss the computational image processing methods, including digital Fourier analysis and machine learning, we employ to tackle this problem.
Differential Dynamic Microscopy Enhanced with Convolutional Neural Networks
Optical microscopy methods are commonly employed to measure rheological properties, like viscosity or viscoelasticity, of materials from biological tissues to colloidal suspensions. Such rheological properties are important for applications such as formulating pharmaceutical ointments to tailoring the mouthfeel of foods and for answering many fundamental questions such as how phase separation proceeds and how the material properties of the cytoplasm respond to stimuli. In recent years, an optical microscopy technique for performing rheological measurements known as differential dynamic microscopy (DDM) has grown in popularity. However, a drawback of DDM is that one typically needs to acquire movies of ~1000 frames of a sample to quantify that sample's properties. If multiple samples or multiple locations within a sample need studying, this is a time-consuming process. Therefore, we devised a method to record a movie of our sample as we scan the sample using a motorized stage. This allows us to measure properties across a large region of the sample (or possibly multiple samples) quickly. However, because the sample is moving as we record images, the computational image analysis becomes more complicated. We discuss the computational image processing methods, including digital Fourier analysis and machine learning, we employ to tackle this problem.