| Project Development Time Reduced by 60 percent
Current statistics show that one in 20 women in Singapore has breast cancer. There are about 700 to 1000 new cases diagnoses each year, and 250 women die from the disease. (Source: Breast Cancer Foundation, Singapore).
Mammograpic screening is widely used for early detection of breast cancer, however it comes with a high false-positive rate. According to a study by the National Cancer Institute USA, 80% of American women who undergo surgical breast biopsies do not have cancer.
Having taken both the traumatic nature and cost of biopsy into consideration, under the supervision of Associate Professor Er Meng Joo, Mr Lim Wei Keat, a 25-year-old research student studying in the Nanyang Technological University of Singapore (NTU) embarked on a project ‘Classification of Breast Masses Using Dynamic Fuzzy Neural Networks’ to develop computer-based methods to distinguish accurately between benign masses and malignant tumours.The Challenge In order to reduce unnecessary biopsies caused by false-positive interpretations, Wei Keat set out to develop an intelligent system, which serves to provide a ‘second opinion’ to radiologists and also to demonstrate the feasibility of using Dynamic Fuzzy Neural Networks (DFNN) in classifying breast cancer.
“With this project, I would like to come up with a system which will be able to extract rules so that radiologists will be more confident in using the system and perceive its reasoning,” said Wei Keat.The Solution Wei Keat chose to work with MATLAB and its related toolboxes (Image Processing, Neural Network and Fuzzy Logic Toolboxes) because they provided an environment that enabled precise, convenient and most importantly, rapid development of the system.
By using MATLAB, it allowed Wei Keat to concentrate more on developing his ideas and algorithms rather than on lengthy programming.
“MATLAB is easy to learn and there isn’t a need to programme a lot of functions as most of them are already available, and it has helped me save plenty of time. It took me one-third of the time to complete my programming as compared to if I had used C,” said Wei Keat.
There are two stages in this computer-aided diagnosis system. In the first stage, as texture features had to be extracted from the mammogram, MATLAB and the Image Processing toolbox were used. Not only do they perform well in manipulating big datasets, images could also be converted into matrices through image processing techniques easily.
In the second stage, MATLAB was used to program human knowledge of learning and reasoning into the computer system so that it would be able to recognise and classify tumours through examples using dynamic fuzzy neural networks.
 
“Compared with other software like C, MATLAB is very good in manipulating large amounts of data and doing image processing is much easier. It definitely cut short my programming time with ease.”
Lim Wei Keat
Finalist in FEER’s Young Inventor’s Award 2002
Research Student
School of Electrical and Electronic Engineering
Nanyang Technology University
|
Results
· Based on a total of 77 cases from the Digital Database of Screening Mammography which were analysed there appeared to be better and more accurate results with the computer-aided systems
|
TP = true positive, FP = false positive
TN = true negative, FN = false negative
Accuracy = (TP+TN)/(TP+TN+FP+FN) = 84.4%
Positive predictive value = TP/(TP+FP) = 78.7%
Negative predictive value = TN/(TN+FN) = 93.3%
|
· Rules could also be extracted from the black box to enable people to see how the system works
· This software is also compatible with both types of mammography, the conventional
X-ray and the digital mammogram.
“We’re really satisfied with the results so far, however, we’re still trying to improve, we won’t stop just here, hopefully with more cases, we can fine-tune the system with the help of MATLAB, to come up with better performances,” enthused Wei Keat.
With the completion of this project, breast cancer can now be detected with little increase in cost as compared to paying for double readings by two radiologists. This method will also reduce unnecessary biopsies caused by false-positive interpretations and help in the selection of appropriate treatment.
Special Note
Wei Keat emerged a finalist in the Far Eastern Economic Review’s Young Inventors Awards 2002 with this project.
“The project attracted a large crowd during the final of the FEER Young Inventor’s Awards 2002. I am confident that the product will do well in the market as it is low-cost and is more accurate than many state-of-the-art methods,” said Professor Er.
Wei Keat is currently awaiting more data from local hospitals so as to carry out clinical trials at the local hospitals. Dr Mark Brooks, an Interventional Radiologist at Austin Repatriation Medical Centre in Australia is also keen to carry out clinical trials with NTU using this new technique.
|
