Gradient-Boosting Algorithm for Microwave Breast Lesion Classification—SAFE Clinical Investigation

Abstract

(1) Background: Microwave breast imaging (MBI) is a promising breast-imaging technology that uses harmless electromagnetic waves to radiate the breast and assess its internal structure. It utilizes the difference in dielectric properties of healthy and cancerous tissue, as well as the dielectric difference between different cancerous tissue types to identify anomalies inside the breast and make further clinical predictions. In this study, we evaluate the capability of our upgraded MBI device to provide breast tissue pathology. (2) Methods: Only patients who were due to undergo biopsy were included in the study. A machine learning (ML) approach, namely Gradient Boosting, was used to understand information from the frequency spectrum, collected via SAFE, and provide breast tissue pathology. (3) Results: A total of 54 patients were involved in the study: 29 of them had benign and 25 had malignant findings. SAFE acquired 20 true-positive, 24 true-negative, 4 false-positive and 4 false-negative findings, achieving the sensitivity, specificity and accuracy of 80%, 83% and 81%, respectively. (4) Conclusions: The use of harmless tissue radiation indicates that SAFE can be used to provide the breast pathology of women of any age without safety restrictions. Results indicate that SAFE is capable of providing breast pathology at a high rate, encouraging further clinical investigations.

Review and Analysis of Tumour Detection and Image Quality Analysis in Experimental Breast Microwave Sensing

“…Half of the research in image-based patient diagnosis has been performed using subjective diagnostic criteria [49,111,141,143,170]. These articles relied on subjective judgments of diagnosticians to determine the presence of a cancerous response in the reconstructed image.…”

“…Subjective diagnostic criteria have been commonly used. [100,111,120,141,143,170,201], subjective diagnosis [49,111,141,143,170], and sample size limitations were common [48,143,201].…”

“…These studies are mentioned here for completeness as there are relatively few applications of machine learning for tumour detection within breast microwave sensing. Other studies, which classified known lesions in patient data [48,49,93,196], were included due to their use of patient data.…”

“…The studies that were selected for possible inclusion based on their abstract and title (n = 468) were then evaluated with respect to the inclusion and exclusion criteria. One hundred seventy-five papers were identified for inclusion in this review [11][12][13][14][24][25][26][27][28][29][30][31][33][34][35][36][37][38][39][40][41][42][43][44][45][46][47][48][49][51][52][53][54][55][56][57][58][59][60][61][62][63][64][65][66][68][69][70][71][181]…”

“…Current challenges in image-based estimation of the diagnostic performance of BMS. Nonblind studies[100,111,120,141,143,170,201], subjective diagnosis[49,111,141,143,170], and sample size limitations were common[48,143,201].…”