AI software test shows promise for breast ultrasound diagnoses
A test of artificial intelligence in assessing breast ultrasound examinations has offered some evidence that it aids in cancer detection and boosts physician accuracy.
A test of artificial intelligence in assessing breast ultrasound examinations has offered some evidence that it aids in cancer detection and boosts physician accuracy.
The test, conducted by a New York-based private practice radiology group that performs a high volume of such exams found promising results in assessing lesions found during breast ultrasound examinations.
The evaluation spanned 18 months and involved nine specialist radiologists who used artificial intelligence software from Koios Medical, which has been cleared for use by the Food and Drug Administration.
The company did not release the name of the radiology practice.
The software was used to provide second opinions, and the assessment found that cancer detection rates increased; at the same time, it reduced false positive biopsy rates as much as 20 percent during the 18 months of the test.
During the evaluation period, the radiologists analyzed more than 6,000 diagnostic breast ultrasound exams, then used Koios DS Breast decision support software to assist in lesion classification and risk assessment. The software, integrated into the radiologists’ normal diagnostic workflow, would provide findings that the radiologists could review in tandem with clinical details to formulate suggestions for case management.
Researchers analyzed the physicians' diagnostic performance during the 18-month test to how they fared during the 18 months prior to the introduction of the AI enabled software. When using the software, physicians recommended biopsy for suspicious lesions at a similar rate (17 percent) and performed 14 percent more biopsies, increasing the cancer detection rate (from 8.5 to 11.8 per 1,000 diagnostic exams).
At the same time, there was a significant reduction in benign biopsies, otherwise known as false positives. Trailing six-month results indicate a benign biopsy reduction exceeding 20 percent across the group. Positive predictive value—the percentage a positive test returns a positive result—improved more than 20 percent.
Koios DS Breast 2.0 is artificial intelligence software designed around a dataset of more than 450,000 breast ultrasound images with known results intended for use to assist physicians analyzing breast ultrasound images and aligns a machine learning-generated probability of malignancy. This probability is then checked against and aligned to the lesion's assigned BI-RADS category, the scale physicians use to recommend care pathways.
The test, conducted by a New York-based private practice radiology group that performs a high volume of such exams found promising results in assessing lesions found during breast ultrasound examinations.
The evaluation spanned 18 months and involved nine specialist radiologists who used artificial intelligence software from Koios Medical, which has been cleared for use by the Food and Drug Administration.
The company did not release the name of the radiology practice.
The software was used to provide second opinions, and the assessment found that cancer detection rates increased; at the same time, it reduced false positive biopsy rates as much as 20 percent during the 18 months of the test.
During the evaluation period, the radiologists analyzed more than 6,000 diagnostic breast ultrasound exams, then used Koios DS Breast decision support software to assist in lesion classification and risk assessment. The software, integrated into the radiologists’ normal diagnostic workflow, would provide findings that the radiologists could review in tandem with clinical details to formulate suggestions for case management.
Researchers analyzed the physicians' diagnostic performance during the 18-month test to how they fared during the 18 months prior to the introduction of the AI enabled software. When using the software, physicians recommended biopsy for suspicious lesions at a similar rate (17 percent) and performed 14 percent more biopsies, increasing the cancer detection rate (from 8.5 to 11.8 per 1,000 diagnostic exams).
At the same time, there was a significant reduction in benign biopsies, otherwise known as false positives. Trailing six-month results indicate a benign biopsy reduction exceeding 20 percent across the group. Positive predictive value—the percentage a positive test returns a positive result—improved more than 20 percent.
Koios DS Breast 2.0 is artificial intelligence software designed around a dataset of more than 450,000 breast ultrasound images with known results intended for use to assist physicians analyzing breast ultrasound images and aligns a machine learning-generated probability of malignancy. This probability is then checked against and aligned to the lesion's assigned BI-RADS category, the scale physicians use to recommend care pathways.
More for you
Loading data for hdm_tax_topic #better-outcomes...