Fuzzy logic can be used to quickly and accurately identify the tell-tale signs of COVID-19 in lung scans and X-rays of patients suspected of having the disease, according to new work published International Journal of Intelligent Information and Database Systems.
Fariha Noor, Md. Rashad Tanjim, Muhammad Jawadur Rahim, Md. Naimul Islam Suvon, Faria Karim Porna, Shabbir Ahmed, Md. Abdullah Al Kaioum, and Rashedur M. Rahman of North South University, in Dhaka, Bangladesh, explain that image processing is crucial in many areas of scientific and medical investigation. This is no truer than with respect to determining whether a patient presents with COVID-19, no infection, or unrelated viral pneumonia.
The team has used two approaches to segmenting images – fuzzy c-means, and k-means clustering. This allowed them to map out the key features of computerized tomography (CT) images and X-rays from known patients with a diagnosis and then use the data to train a convolutional neural network to identify the characteristics in new images presented to it. As they hoped, the approach worked much better with segmented images than with raw images. Moreover, both CT and X-ray images gave good results. The team adds that they could improve accuracy still further when they also applied fuzzy edge detection to the images.
The team adds that there is plenty of room for improvement in the accuracy of the approach but suggests that optimization and the classification of greater numbers of images will allow this to happen quickly. For any convolutional neural network, the more classified data, i.e. known images with which it is trained, the better in terms of boosting accuracy and reducing the likelihood of false positive or false negative results from the diagnostic. The researchers also suggest that the same approach might also be used to classify other diseases.
In conventional Boolean logic, a variable can only be binary, toggling between 0 and 1, false or true. In fuzzy logic, invented in the mid-1960s, there is a suggestion that a result can be on a spectrum, and so have a non-integral value lying between 0 and 1. The allusion being that an output can lie somewhere between completely false and completely true. Now, this is not to suggest that a disease diagnosis might be a half-truth. Rather, where there is ambiguity in the data or results, in this case, CT or X-ray images, a secure decision can be made by a neural network, for instance, about the nature of each segment of the image being associated with infection or otherwise. When multiple segments are then investigated, a more likely diagnosis of negative or positive can be gleaned from the images based on how well the system has been trained with definitive images.
Noor, F., Tanjim, M.R., Rahim, M.J., Suvon, M.N.I., Porna, F.K., Ahmed, S., Al Kaioum, M.A. and Rahman, R.M. (2021) ‘Application of fuzzy logic on CT-scan images of COVID-19 patients’, Int. J. Intelligent Information and Database Systems, Vol. 14, No. 4, pp.333–348.
