Manual inspection of cellular assays for QC purposes is regularly a bottle-neck in production, and data repeatability is almost impossible. Several systems now exist on the market offering automated analysis of cellular forms but poor reliability, data quality and validation issues often mean the benefits are not realised. Successful automation of cellular analysis can however be reliably achieved through ‘Vision Assisted Inspection’, a step-wise de-risked approach to productivity improvement and data accuracy, previously used by PA to provide orders of magnitude improvement to our clients.
Automated image processing solutions usually require explicit mathematical models or rules to define how objects are to be classified. Biological features tend not to conform, but they do have:
Complex features to quantify
No explicit mathematical rules for decision making
No explicit boundaries or thresholds between categories
Continuous distribution of features across categories
A degree of human expert interpretation is required. Therefore attempts to fully automate such counting tasks often fall short. The concept of vision assisted inspection systems has been developed to encompass these factors. It aims to work within the constraints to automate only those parts of a process that can be handled robustly.
Advances in algorithms and computer power mean automation can now handle a very large part of the process. But by recognising from the outset that there will be borderline areas where human decision making will still be required, the machine assistance in those areas takes the form of a user-friendly interface to present data to the operator for classification.
PA’s approach to developing machine assisted counting systems proceeds through a number of phases. Each phase of development is designed to maximise benefits while minimising costs and mitigating risks and to work in parallel with existing manual processes until the new technology has been validated. The methodology can also be applied to other areas, for instance assessing the surface finish of medical devices such as stents.
PA has used this approach in a number of research, drug discovery and production environments. Past benefits include:
a 7-fold increase in data repeatability
a 5-fold improvement in accuracy of results
in one example, throughput almost doubled while releasing 20 people to do other work.
To find out more about PA's approaches to improving productivity, please contact us now.