ANALYSIS OF THE RESULTS OF BREAST NEOPLASM STRATIFICATION BASED ON

ULTRASOUND DATA USING ARTIFICIAL INTELLIGENCE: IS IT WORTH RELYING

ON THE "MACHINE OPINION"?

 

Marushchak E.A.^1,2, Fisenko E.P.¹, Zubareva E.A.²

 

1 - B.V. Petrovsky Russian Research Surgery Center. Moscow, Russia.

2 - N.I. Pirogov Russian National Research Medical University. Moscow, Russia.

P

urpose. To provide the comparative analysis of the discrepancy between the results of the stratification of breast neoplasms according to ultrasound examination using artificial intelligence software.

Materials and methods. In 2024 in the department of ultrasound diagnostic of the SCC hospital No. 2 of B.V. Petrovsky Russian Research Surgery Center 91 multiparametric ultrasounds were performed (B-mode, CDK, elastography, assessment of microvascularization), in patients of 18-70 years old with 129 breast tumors, including using the Samsung RS85 S-Detect AI ultrasound scanner software. Then, morphological verification of neoplasms of various levels was performed in all patients (fine needle aspiration biopsy, trepan biopsy, vacuum aspiration biopsy, total histological examination with immunohistochemical analysis).

Results. A comparative analysis of the results of stratification of breast neoplasms based on ultrasound data was conducted with and without the use of artificial intelligence programs. The coincidence of the AI data and the ultrasound doctor was in 117 (91%) neoplasms, data discrepancies were in 12 neoplasms (9%). The coincidence of the doctor's ultrasound and morphological verification data was 90%, and the AI – 87%. The sensitivity of ultrasound performed by a doctor and AI turned out to be the same and amounted to 82%; the specificity of ultrasound performed by a doctor was 92%, AI – 88%. At the same time, the number of true-positive cases of stratification in doctor was 9, true-negative – 108; false-positive – 10, false-negative – 2. Using AI, the following data were obtained: the number of true-positive cases – 9, true-negative – 104, false-positive – 14, false-negative – 2. This article analyzes the discrepancies between the stratification data of breast neoplasms according to the BI-RADS (Breast Imaging-Reporting and Data System) scale of the doctor and the AI in 12 patients, as well as the analysis of the discrepancy between the stratification data based on the results of a morphological study with the concurrence of the opinions of the doctor and the AI (8 patients).

Discussion. When analyzing cases of discrepancies between AI and specialist data, as well as the coincidence of AI and specialist data, but discrepancies with the results of morphological examination, it was revealed that the decisive factors leading to AI errors are the fuzzy contours of the neoplasm, and the inability to analyze the totality of the patient's medical data. AI, which currently does not have sufficiently advanced algorithms similar to the specialist’s clinical thinking, forms erroneous conclusions in cases other than the "standard" ones. An important point when performing ultrasound using AI software is the clear positioning of the survey window on the object to be analyzed, as well as good image quality.

Conclusion. The AI ultrasound scanner software can be used in the practice of an ultrasound diagnostics doctor as a decision support system, but taking into account the imperfections of algorithms that exist today, which are influenced by a number of factors that potentially lead to erroneous interpretation of data. The most promising niche for using AI at the moment is screening studies in conditions of limited medical resources. The most significant limiting factors of the rapid progressive development of machine learning are operator-dependent identification of the area of interest, limited high-quality digital data, as well as still imperfect algorithms for analyzing radiomics data.

 

Keywords: ultrasound diagnostics, mammary gland, breast, artificial intelligence, AI, morphological verification, machine learning, deep learning, BI-RADS.

 

Corresponding author: Marushchak E.A., е-mail: Этот e-mail адрес защищен от спам-ботов, для его просмотра у Вас должен быть включен Javascript

 

For citation: Marushchak E.A., Fisenko E.P1, Zubareva E.A. Analysis of the results of breast neoplasm stratification based on ultrasound data using artificial intelligence: is it worth relying on the "machine opinion"?. REJR 2025; 15(3):126-141. DOI: 10.21569/2222-7415-2025-15-3-126-141.

Received:        19.05.25                 Accepted: 30.10.25