Business Systems and Knowledge Modelling Lab

Research Project

Unstained Human Cell Image Processing Technology

Research Goal
Our society demands rapid turn over rates of diagnosed samples in the medical field. In the majority of cases, the rapidity of specimen analysis determines when a patients medical treatment begins. With our growing population there will be an increased demand for automatic diagnostic equipment to cater for the greater number of specimens being processed and analysed in a cost effective manner. Most automated diagnostic systems for detection of cervical cancer cells (e.g., Lerma et al., Cancer, Vol. 84(6), (1998)361) by pap smear, detection of external skin melanoma, detection of prostate cancer, detection of plaques in Alzheimer’s disease are based on use of stained specimens.

The staining method used in the slide preparation process has been the accepted method for many years in the laboratories as it can increase the contrast of cell border and the background. However, it involves three to four hours preprocessing of cell samples. The staining process can add chemical effects to the nature of the cell, introduce changes in the living cells functionality after the stained cell sample has been left for a while. Therefore there is a need for developing a new technology for detection of unstained cells rather than using present staining technology. There is also a commercial market of $10 million dollars for unstained cell imaging technology The existing unstained human cell imaging technologies have classification accuracy in the range of 40 to 45%.

A part of the VPAC expertise grant from round 1 was used for investigating the design of this new technology. Subsequently we also received a small ARC grant on this project. The existing research results have been publication. In contrast to existing approaches where image segmentation and classification have been treated as two separate processes we have designed a control system image processing model in which the image segmentation and classification are treated as one process.

Funding
$70,000

Acknowledgements
Auto Scan Ltd., Melbourne

Team Members
Auto Scan Ltd, Rajiv Khosla, Chris Lai, Jamie Gadd, Yasue Mitsukura.

Publications
[1] R. Khosla, C. Lai and Y. Mitsukura, “Human-Centered Multi-Agent Distributed Architecture for Knowledge Engineering of Image Processing Applications,” to appear in the International Journal of Pattern Recognition and Artificial Intelligence Volume 18 Number 1 February 2004.

[3]. R. Khosla, C. Lai and Y. Mitsukura, “ Optimising the Performance of Soft Computing Agents for Classification of Unstained Mammalian Cell Images,” to appear in IEEE 2003 International Symposium on Computational Intelligence for Measurement Systems and Applications, IEEE Computer Society, Lugano, Switzerland, pp. 199-204, July 2003.

[4] C. Lai and R. Khosla, “Genetic Algorithm Based Optimisation of a Multi-Agent Soft Computing Model for Segmentation and Classification of Unstained Mammalian Cell Images, in IEEE International Conference on Evolutionary Computation, Canberra, Australia, pp. 1119-27, December 2003

[5] E. Damiani and R.Khosla, Human-Centered Approach to Medical Imaging Systems, in 6th International conference on Fuzzy Days, Lecture Notes in Computer Science, Springer-Verlag, Dortmund, Germany, pp. 1999