Dielectric Characterization of Dispersive Head Tissue for Detection and Classification of Tumour Using Microwave Imaging Technique and Deep Learning Model

• Published in: Arabian journal for science and engineering (2011) Vol. 49; no. 9; pp. 12305 - 12316
• Main Authors: K, Lalitha, J, Manjula
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 2024; Springer Nature B.V
• Subjects: Algorithms; Artificial neural networks; Brain; Brain cancer; Data acquisition; Data augmentation; Datasets; Engineering; Humanities and Social Sciences; Image acquisition; Image reconstruction; Imaging techniques; Machine learning; Medical imaging; Microwave antennas; multidisciplinary; Object recognition; Parameter modification; Parasitic elements (antennas); Radiators; Research Article-Computer Engineering and Computer Science; S parameters; Science; Signal classification; Tissues; Tumors; Antipodal Vivaldi antenna; Biomedical electromagnetic imaging; Brain tumour; Inverse scattering; Microwave head imaging
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-023-08666-z

Microwave sensing is envisioned to be a reliable imaging technique for detecting and monitoring brain tumours. It uses low-power, nonionized microwave signals to visualize the interior of the living tissues. This study automatically exploits categorization of brain tumours using a deep neural network called the YOLOv5 l algorithm in non-invasive microwave images. A modified antipodal structure (including parasitic patch between flares) that operates over the wideband is proposed for the collection of scattering parameters. The staircase parasitic patch located at the centre of the radiator provides the improved radiation performance at low frequencies. Subsequently, a simulation environment was modelled using CST Microwave Studio 2020, which contained a data acquisition system using a designed microwave antenna and head phantom. Next, the image reconstruction is implemented MATLAB software. The detection and classification of tumours were performed using the YOLOv5l model. A dataset with 500 image samples was formed by pre-processing, which was divided into training (70%), testing (15%) and validation (15%). Then, augmentation is used to create final dataset with 1000 training images. The model was successful in locating an early-stage brain tumour with a size of 5 mm. The YOLOv5 model automatically detected tumour (s) and categorized them as benign tumour or malignant tumour based on a predicted bounding box with an accuracy of 91.5%.