Siberian Federal University Scientists System To Detect Breast Tumor Boundaries With Newly Developed System
An Iraqi-Russian team of scientists has developed a system for detecting breast tumor boundaries. The new system will automate tumor detection and segmentation and help radiologists significantly reduce the diagnostic time and increase the accuracy of determining infected and healthy areas.
Breast tumor segmentation and boundary detection are crucial stages in breast cancer therapy and patient follow-up. The scientists’ research is aimed at creating a system for accurately segmenting breast tumors and non-infected areas of the breast on medical imaging using a combination of Fuzzy Clustering Means and Threshold (FCMT). The computer-aided diagnostic method works on each breast slice without any training for segmentation and boundary detection.
“The main tasks of our system are segmentation, selection of boundaries and measurement of the size of breast tumors. The product is now being tested and finalized,” informed Yousif Ahmed Hamad, co-author of work and research engineer at the Artificial Intelligence Laboratory, SibFU.
The system consists of several stages. The input data is a breast mammogram used to diagnose tumors and breast cancer. A medical image is converted to grayscale if it is presented in RGB format. Next, the image is scaled to the appropriate matrix to maintain the aspect ratio of the image. After preparation, the resized image is subjected to a median filter that minimizes random noise while maintaining its specified image boundaries.
“A noise removal technique is used to improve the quality and contrast when improving the original image scanning. To augment and isolate the area of foreign bodies (tumors or nodules), we used the balance contrast enhancement technique (BCET). Segmentation and measurement of the medical image is recommended after image augmentation for the accurate determination of the infected area boundaries. For segmentation, we used a combination of FCM and threshold. The threshold is needed to convert the filtered image to binary one to highlight the object of study in the breast image. FCM is used for segmentation of the infected area of the breast (tumor). The last stage of the study is the Canny edge detector which clearly detects healthy areas of the gland and tumors based on the developed segmentation method,” explained Anastasiia Safonova, head of the research, associate professor at the SibFU’s Department of Artificial Intelligence Systems.
A comparison of the new algorithm with widely used neural network algorithms (Sonet and UNet) showed that the prediction accuracy of the new product is 18% higher.
According to the developers, similar algorithms are already used in medicine, but today they are considered an auxiliary method — they help the doctor diagnose and detail the tumor boundaries, but they do not completely replace expert opinion. The scientists also noted that the proposed algorithm can be adapted, among other things, to identify various lung pathologies, both with minor modifications, and in an already existing form.