Answers are grounded in this paper and cite their sources.
Mammographic breast density is an important diagnostic indicator. An increase in breast density reduces diagnostic accuracy; moreover, as density increases, the risk of breast cancer also rises. It is worth noting that not only high density can indicate possible pathological changes. Low density can be a marker of increased risk of cardiovascular diseases (CVD). Based on recent studies conducted worldwide, there is evidence of an increased risk of arterial hypertension, ischemic heart disease, heart failure, cerebrovascular disease, hypercholesterolemia, as well as diabetes mellitus. Aim of the reseach: to assess the agreement among three AI services: AI-service No 1, AI-service No 2, AI-service No 3 according to the ACR density scale. Material and methods. A mixed study was conducted, combining retrospective diagnostic analysis and analytical research. Based on data from ERIS EMIAS, 99 anonymized mammographic studies of women over 18 years old (mean age 56 years) were selected for the period from 13.11.2020 to 04.10.2021, excluding studies with artifacts on the images. Each study was independently evaluated by an expert physician (a radiologist with ≥ 5 years of experience and an academic degree) and three AI services (AI-service No 1, AI-service No 2, AI-service No 3) with determination of breast density according to the ACR BI-RADS scale. The main objective of the study was to assess the agreement between the expert and AI services, which was analyzed using the intraclass correlation method (Pearson). Results. In our study, the diagnostic accuracy parameters of AI services determining breast density according to the binary ACR scale were as follows: ROC AUC – 0.866–0.904, sensitivity – 0.833–0.867, specificity – 0.899–0.957, accuracy – 0.879–0.919. When assessing diagnostic accuracy for individual breast density categories according to the full ACR scale, the following values were obtained: ROC AUC – 0.817–0.995, sensitivity – 0.714–1.000, specificity – 0.784–1.000, accuracy – 0.828–0.990. Conclusion. All AI-services demonstrated high sensitivity (0.833–0.867) and specificity (0.899–0.957), which is critically important for screening tools aimed at minimizing both false-negative and false-positive results.
No takes yet. Share an insight, caveat, or question.
Khrustacheva et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: