Benign breast lesions : analysis by artificial intelligence and removal by vacuum-assisted excision

Following advances in diagnostic imaging modalities over the last decade, lesions of uncertain malignant potential have been increasingly diagnosed, primarily through mammography and sonography. These diagnoses, however, increase patient anxiety and result in an abundance of work-ups, including biopsies for radiologists, and often lead to unnecessary surgery.

One modern field of research is the integration of artificial intelligence (AI) based computer-aided detection (CAD) systems into various types of equipment to improve their accuracy. Our first study (Study 1) investigated the grading of previous benign biopsies using an AI-CAD system that has been integrated into mammography.

Another modern field of research has been the use of larger bore needles in breast biopsies, such as vacuum-assisted biopsy needles with an outer diameter of up to 7G (4.6 mm). These needles can provide a sufficient tissue sample with which to obtain a more accurate diagnosis, while at the same time allowing the operator to completely excise the specimen under local anesthesia. Our second study (Study 2) investigated how needle size affected the time and results of the excision procedure. Our third study (Study 3) evaluated the procedure from the patients' perspective, documenting their experiences and any eventual adverse effects after the procedure. Our fourth study (Study 4) compared the first diagnostic pathology report to the reports obtained post-excision to identify lesion characteristics that would help determine which lesions are more susceptible to excision with a larger needle.

In Study 1 we retrospectively applied a commercial AI-CAD system (Insight MMG, version 1.1.4.3; Lunit Inc.) to a dataset of screening mammograms from 10,889 women. We divided the study population into three groups: women who did not undergo a biopsy, those who underwent a biopsy before or after screening mammography (with benign results), and those who were diagnosed with breast cancer. The AI system flagged all women above the cutoff threshold, which was defined as 0.4 on a scale of 0.0 to 1.0. The percentages of women flagged were as follows: 3.5% for healthy women without a biopsy, 11% for those with benign biopsy findings, and 84% for those with breast cancer (P < 0.001). The AI-CAD system flagged a much larger proportion of women who underwent a biopsy than those who did not; however, the flagging rate was not any higher than that of the radiologists.

In Study 2 we performed a randomized controlled trial to compare the excision completeness and efficacy of the vacuum-assisted excision (VAE) procedure using 7G and 10G vacuum needles. We enrolled 208 patients, and after withdrawal of consent, the trial population included 194 patients. There were no differences in procedure time (P = 0.126) or excision completeness (P = 0.109) between procedures performed using 7G and 10G needles. Of the 127 patients who attended the 24-month follow-up, 88% (112/127) had lesions completely excised, with no statistically significant difference between the 7G and 10G needles.

In Study 3 we administered a questionnaire to all of the patients included in Study 2. Patient acceptance of the procedure and short- and long-term complications were also documented. We calculated the total hospital costs of the VAE procedures and compared them with those of open surgical excision (OSE), the previous standard of care for surgical excision. There were no significant differences in pain levels (P = 0.713), complications (P = 0.724), or patient acceptance of the procedure between the 7G and 10G needle groups (P = 0.401). Approximately 97% (173/178) of the patients would recommend the procedure to others, and the total hospital procedural cost of VAE was estimated to be 60% lower than that of OSE.

In Study 4 we retrospectively examined the results of the pathology reports of all patients included in Study 2; however, we excluded patients who did not have a cytological or histopathological diagnosis prior to the VAE, during which tissue samples were placed in one, two, or three successive containers, starting at the core of the lesion and moving outwards to the normal tissue. The results of the diagnostic reports from the initial biopsy (cytology and/or histology) were compared with those from the tissues obtained during the VAE. The discrepancy between the diagnoses of fine needle aspiration (FNA) specimens and those from VAE was 38%, while that for core needle biopsy (CNB) was 29%. The upgrade rate to cancer was most common after a diagnosis of atypical ductal hyperplasia (ADH) on CNB.

In conclusion, this thesis provides new knowledge on how to improve the performance of AI-CAD systems and broadens our understanding of we can improve the performance of AI. This confirms the necessity for alternative solutions to surgery for the diagnosis and treatment of undetermined lesions and provides data to support a separate personalized approach for different lesion types.

List of scientific papers

I. Athanasios Zouzos, Aleksandra Milovanovic, Karin Dembrower, Fredrik Strand

Effect of Benign Biopsy Findings on an Artificial Intelligence-Based Cancer Detector in Screening Mammography: Retrospective Case-Control Study

JMIR AI 2023 | vol. 2 | e48123 | p. 7
https://doi.org/10.2196/48123

II. Athanasios Zouzos, Irma Fredriksson, Andreas Karakatsanis, Iliana Aristokleous, Theodoros Foukakis, Fredrik Strand

Effect of needle size on outcomes of vacuum-assisted excision of breast lesions. A randomized controlled trial

European Journal of Radiology 183 (2025) 111895
https://doi.org/10.1016/j.ejrad.2024.111895

III. Athanasios Zouzos, Irma Fredriksson, Andreas Karakatsanis, Fredrik Strand

Patient experience and healthcare cost aspects of vacuum-assisted excision of breast lesions. A report from the Swedish VAE randomized clinical trial

[Manuscript]

IV. Athanasios Zouzos, Irma Fredriksson, Andreas Karakatsanis, Johan Hartman, Fredrik Strand

Variation in pathological appearance across repeated sampling from probably benign breast lesions

[Manuscript]