Computer-aided detection (CAD) software developer CureMetrix reported in a Tuesday morning presentation at the virtual RSNA 2020 meeting that radiologists who used its mammography artificial intelligence (AI) software in a recent challenge were more accurate than those who didn't utilize the technology.
In a "Test Drive Challenge" study performed by the company in collaboration with Ambra Health, CureMetrix found that radiologists who used its AI software had improved sensitivity and specificity for detecting breast cancer compared with those who didn't. They also read normal cases much faster than those who interpreted the exams without the aid of AI software, said Kevin Harris, president of CureMetrix.
In the study, CureMetrix first invited seven experienced radiologists to read an enriched test of 25 mammograms, which included 19 normal and six biopsy-confirmed cancer cases, according to the vendor. These exams were interpreted without the use of AI on a research platform created by Ambra, according to the vendor.
Next, 30 other radiologists who volunteered to participate in the challenge read the same cases on the same platform, but with help from AI. In this part of the study, the radiologists were given a worklist of the 25 mammograms that had been prioritized by the firm's cmTriage mammography triage software by level of cancer suspicion. They then interpreted the exams with help from the vendor's cmAssist algorithm, which identifies, marks, and scores suspicious lesions for review by the radiologist, according to CureMetrix.
| Radiologist performance on test set of 25 screening mammograms | |||
| Radiologists without AI | Radiologists with AI | Percentage improvement | |
| Sensitivity | 71% | 75% | 5% |
| Specificity | 56% | 75% | 34% |
| Accuracy | 60% | 75% | 25% |
The improvements were seen across all levels of reader experience, according to the company.
In addition, the radiologists were able to read normal exams 40% faster with use of AI, CureMetrix said.
