Size Based Characterization of Gold Nano Particles using Machine Learning Approach

In drug delivery, there is a need for precision in reporting particles parameters. Studies have shown that absorbency of drugs in the blood stream depends on the size of the nanoparticle. The shape and size of nanoparticles (NPs) matter the most, hence the distribution of NP depends on the size and shape of NPs. By synthesizing and characterizing the NPs, we are able to cluster and get the amount of a certain type of morphology and accurate size determination. Moreover, the size distribution of a particle plays a more important goal as it possesses an increase in the usability of a diagnostic and therapeutic tool in medicine. The shape and size distribution of NPs is important for the delivery of drugs and for the cure or treatment of several chronic diseases such as cancer. Hence it is important to get the accurate size distribution of NPs for better results. Gold nano particles (AuNPs) where measured manually by the use of transmission electron microscope, hence, in most cases human error could play part in terms of inaccurate measurements. The digital images of AnNPs contain noise, making it difficult to get accurate measurements using the transmission microscope. AuNPs were measured in terms of their width and length. This study focused on the characterization of AuNPs collected by the transmission electron microscope using machine learning approaches. Image preprocessing and processing techniques are used for extracting the features (length and width) of AuNPs. In this study, filtering techniques such as Gaussian blur, Median and Mean filtering techniques are employed for noise removal to increase the precision in estimating the size of NPs. Unsupervised machine learning algorithm such as K-means and Otsu are used to perform image segmentation of the filtered nano images for the accurate extraction of particles' features such as length and width. The size measurements obtained using the machine learning approaches are compared with the measurements taken by the transmission electron microscope (TEM) for error estimation in the size distributions of NPs. The results showed that machine learning approaches provided accurate measurements of most of the NPs as compared to TEM. Therefore, it is recommended that machine learning approaches can be used to estimate the size of NPs so that the shapes can be described better and classified during the synthesis process.