By facilitating more effective triage of patients for breast MRI, an emerging artificial intelligence (AI) software that reportedly enhances assessment of breast density may bolster the detection of invasive breast cancer.
For the prospective randomized clinical trial, recently published in Nature Medicine, researchers assessed the use of AISmartDensity for triage of patients for supplemental breast MRI after negative mammography exams. The study authors said the AI-powered software incorporates convolutional neural network recognition of inherent breast cancer risk, masking potential and cancer signs trained on mammography images.
The cohort was comprised of 559 patients (median age of 56), including 22 patients with previous breast cancer and 104 patients who had a family member with breast cancer history, according to the study. Breast MRI exams, assessed by two radiologists with four and five years of experience, respectively, revealed 54 BI-RADS 3 lesions, 27 BI-RADS 4 lesions and 14 BI-RADS 5 lesions. The researchers noted that subsequent ultrasound did not show localized lesions for 23 BI-RADS 3 cases or a corresponding lesion for one BI-RADs 4 case.
However, the study authors noted that biopsies of the remaining 71 BI-RADS 3-5 lesions showed breast cancer in 36 cases, which translated to a 50.7 percent positive predictive value (PPV) and a cancer detection rate (CDR) of 64.4 cancers per 1,000 MRI exams.
“The cancer detection rate of our trial at 64 … cancers per 1,000 MRIs corresponds to about 3.8 times higher supplemental cancer detection rate compared with the traditional density method used in the DENSE trialat 16.5 cancers per 1,000 MRIs,” wrote lead study author Mattie Salim, M.D., Ph.D., who is affiliated with the Breast Radiology Unit at Karolinska University Hospital and the Department of Oncology-Pathology at Karolinska Institute in Stockholm, Sweden, and colleagues.
While noting that only 8 percent of the diagnosed cancers were associated with lymph node metastasis, the researchers found that 61 percent of the malignant lesions involved a combination of invasive and ductal carcinoma in situ, 36 percent were larger than 20 mm and 19 percent of the cases revealed multiple lesions on breast MRI.
“It is notable that most of the cancers detected in the population selected for supplementary MRI exhibited invasive features,” added Salim and colleagues.
Three Key Takeaways
- Enhanced cancer detection. The AISmartDensity software facilitated a cancer detection rate (CDR) of 64.4 cancers per 1,000 MRI exams, which is about 3.8 times higher than previous research utilizing a traditional breast density assessment method.
- High positive predictive value. The study found a positive predictive value (PPV) of 50.7% for biopsied lesions categorized as BI-RADS 3-5, indicating a robust performance of the AISmartDensity software in identifying potentially malignant lesions.
- Detection of invasive features. Most cancers detected in the population selected for supplementary MRI exhibited invasive features, with 61% of malignant lesions involving a combination of invasive and ductal carcinoma in situ and 36% being larger than 20 mm. This suggests that the AI software effectively identifies more clinically significant cancers.
The researchers also pointed out that 44 percent of the diagnosed cancers exhibited a higher masking potential risk score than the cancer signs component of the AISmartDensity software. However, they noted that previous retrospective research with the AI software showed that the cancer signs assessment was a significant contributing factor to the overall area under the curve (AUC) of AISmartDensity. The study authors maintained that the combination of the convolutional neural network assessment components facilitate improved use of supplemental breast MRI.
“The potential to pre-emptively detect most cancers by offering MRI to a small proportion of individuals represents an important healthcare value proposition,” emphasized Salim and colleagues.
(Editor’s note: For related content, see “MRI-Based AI Model Shows Promise in Predicting Lymph Node Metastasis with Breast Cancer,” “Can AI Automate BPE Assessment of Dense Breasts on MRI?” and “Mammography and Breast MRI: Is it Time to Evaluate Strategies as Opposed to Modalities?”)
In regard to study limitations, the authors noted the AISmartDensity software was trained on mammography images from one vendor. They acknowledged that the small number of breast cancer diagnoses within the study prevented subgroup analysis with respect to breast cancer characteristics and noted no specific assessment of the AI software’s efficacy in women with a history of cancer.
