The manufacture of In Vitro Fertilization (IVF) needles is subject to the most stringent quality demands. This makes automated inspection challenging due to difficulty in reliably classifying conforming and non-conforming (defective) products due to factors including multidimensional variation of their tip geometry and the lack of an explicit quality standard. In addition, developing an IVF needle image dataset, which broadly contains the visual characteristics of qualified and defective products, is difficult without commissioning large and costly production runs. The most important original contribution of this work is a new solution to investigate and quantify the uncertainty in the quality standard of IVF needles by integrating inter-disciplinary techniques. This work utilizes a low-cost, virtual dataset of synthetic images, generated by the automated photo-realistic rendering of a three-dimensional (3D) parametric model to simulate manufacturing variation. Then, the unknown numerical (critical) quality thresholds are obtained by estimating the relationship between quality response and measurement predictors using an Ordinal Logistic Regression (OLR) algorithm on the synthetic images. The fitted models exhibited increased overall predictive accuracy of up to 11.02% than the machine learning models (available in MATLAB) and could provide objective guidance on classifying specific quality aspects of a product.

Accurately tracking a group of small biological organisms using algorithms to obtain their movement trajectories is essential to biomedical and pharmaceutical research. However, object mis-detection, segmentation errors and overlapped individual trajectories are particularly common issues that restrict the development of automatic multiple small organism tracking research. Extending on previous work, this paper presents an accurate and generalised Multiple Small Biological Organism Tracking System (MSBOTS), whose general feasibility is tested on three types of organisms. Evaluated on zebrafish, Artemia and Daphnia video datasets with a wide variety of imaging conditions, the proposed system exhibited decreased overall Multiple Object Tracking Precision (MOTP) errors of up to 77.59%. Moreover, MSBOTS obtained more reliable tracking trajectories with a decreased standard deviation of up to 47.68 pixels compared with the state-of-the-art idTracker system. This paper also presents a behaviour analysis module to study the locomotive characteristics of individual organisms from the obtained tracking trajectories. The developed MSBOTS with the locomotive analysis module and the tested video datasets are made freely available online for public research use.