A diagnostic software algorithm for morphological detection of lobular and ductal carcinoma on ultrasound

Abstract

Breast cancer remains the most common form of cancer amongst women. This has led to the need to enable early detection by using the latest technology in high frequency ultrasound.The increased interest in ultrasound as a diagnostic tool for breast cancer detection has led to rapid developments in the application thereof.Sonography is a popular tool for physicians, radiologists, sonographers and clinical enthusiasts, however true efficacy, in practice, is limited.This study aims to improve diagnostic proficiency by means of calculated algorithms applied to ultrasound images. Through application thereof, inter-rater variability can be reduced.The first tier of the study would be retrospective quantitative study, with cross sectional study design.The second tier would be a prospective case controlled study.The study will be conducted at private radiology practice in the Pretoria area which serves as dedicated womens’ wellness centre.A small mock study sample was tested during the development of the research protocol.The sample population included 12 random selected suspicious mass lesions reported with ultrasound. A senior application specialist of the company MIPAR assisted in the basic recipe/ algorithm which could be used for image segmentation of a tumour within an ultrasound image.The values of the various intensity means were processed as a data set, as seen below in Table 2.It is evident from the mock study sample that ILC has an intensity range of (24-58) in comparison to IDC which is (54-89). This initial evaluation is sufficient evidence for the study to be used prospectively 2019-2020.This study will examine and analyse the validity and utility of the use of retrospective analysis of ultrasound images by means of segmentation software. This will be used in the development of a prospective algorithm tool similar to CAD, but with improved efficacy by means of targeted lesion segmentation and classification through calculated measurements and standardized algorithms (Table 1). Breast cancer remains the most common form of cancer amongst women. This has led to the need to enable early detection by using the latest technology in high frequency ultrasound. The increased interest in ultrasound as a diagnostic tool for breast cancer detection has led to rapid developments in the application thereof. Sonography is a popular tool for physicians, radiologists, sonographers and clinical enthusiasts, however true efficacy, in practice, is limited. This study aims to improve diagnostic proficiency by means of calculated algorithms applied to ultrasound images. Through application thereof, inter-rater variability can be reduced. The first tier of the study would be retrospective quantitative study, with cross sectional study design. The second tier would be a prospective case controlled study. The study will be conducted at private radiology practice in the Pretoria area which serves as dedicated womens’ wellness centre. A small mock study sample was tested during the development of the research protocol. The sample population included 12 random selected suspicious mass lesions reported with ultrasound. A senior application specialist of the company MIPAR assisted in the basic recipe/ algorithm which could be used for image segmentation of a tumour within an ultrasound image. The values of the various intensity means were processed as a data set, as seen below in Table 2. It is evident from the mock study sample that ILC has an intensity range of (24-58) in comparison to IDC which is (54-89). This initial evaluation is sufficient evidence for the study to be used prospectively 2019-2020. This study will examine and analyse the validity and utility of the use of retrospective analysis of ultrasound images by means of segmentation software. This will be used in the development of a prospective algorithm tool similar to CAD, but with improved efficacy by means of targeted lesion segmentation and classification through calculated measurements and standardized algorithms (Table 1). Table 1.Table 1Dataset findings for mock study sampleAlgorithm 1 N = intensity mean (pixel)DatasetILC - TILC - STLC 136.3966105.7651LC 234.133282.4882LC 358.106987.4093LC 445.8658104.4974LC 534.118195.8381LC 624.700198.8514Algorithm 2Dataset 2IDC – TIDC - STDC 154.6937111.7207DC 271.8258118.6943DC 387.8457102.7707DC 461.362895.9202DC 560.802880.1279DC 689.7359114.0227 Open table in a new tab