Digital infrared thermal imaging system based breast cancer diagnosis using 4D U-Net segmentation

Medical Research field has been taken continuous efforts to develop an efficient method for detecting breast cancer, but the goal has still not yet achieved. To overcome this issue, a 4D U-Net segmentation using digital infrared (IR) thermal imaging system is proposed in this manuscript for the diagnosis of breast cancer (DBC-4D U-Net-DITI). Initially, the digital infrared thermal images are taken from DMR-IR data set as input, and the imageries are pre-processed to maintain local features and compress the dynamic range of image based upon Altered Phase Preserving Dynamic Range Compression (APPDRC) approach by removing the speckle noise. Then, the image segmentation is carried out with the help of 4D U-Net for obtaining the segmented digital infrared thermal image. The 4D U-Net weight parameters are optimized with Glowworm Swarm Optimization Algorithm (GSOA). The segmented regions of digital infrared thermal images are fed to Binarized Spiking Neural Network (BSNN) for classifying the pathology stage as No spread, Early Stage, Localized, Regional and Distant. The proposed approach is executed in MATLAB. The performance of proposed approach attains better accuracy of 39.01%, 28.34%, and 37.45%, better precision of 17.12%, 24.12% and 32.07% when compared to existing approaches like chaotic salp swarm algorithm (CSSA) based segmentation of thermal images for breast cancer identification (DBC-CSSA-DITI), marine-predators-algorithm based segmentation of thermal images for the diagnosis of breast cancer (DBC-MPA-DITI) and diagnosis of breast cancer based upon CNN using thermal im-ageries (DBC-CNN-DITI) respectively.

Automated summary

In order to detect breast cancer more efficiently, a 4D U-Net segmentation method using digital infrared thermal imaging system is proposed in this research. The proposed approach utilizes digital infrared thermal images and applies pre-processing techniques, image segmentation with 4D U-Net, and optimization of weight parameters with Glowworm Swarm Optimization Algorithm. The performance of the proposed approach surpasses existing methods in terms of accuracy and precision.