External Evaluation of 3 Commercial Artificial Intelligence Algorithms for Independent Assessment of Screening Mammograms

Abstract

IMPORTANCE A computer algorithm that performs at or above the level of radiologists in mammography screening assessment could improve the effectiveness of breast cancer screening.OBJECTIVE To perform an external evaluation of 3 commercially available artificial intelligence (AI) computer-aided detection algorithms as independent mammography readers and to assess the screening performance when combined with radiologists. DESIGN, SETTING, AND PARTICIPANTSThis retrospective case-control study was based on a double-reader population-based mammography screening cohort of women screened at an academic hospital in Stockholm, Sweden, from 2008 to 2015. The study included 8805 women aged 40 to 74 years who underwent mammography screening and who did not have implants or prior breast cancer. The study sample included 739 women who were diagnosed as having breast cancer (positive) and a random sample of 8066 healthy controls (negative for breast cancer). MAIN OUTCOMES AND MEASURESPositive follow-up findings were determined by pathology-verified diagnosis at screening or within 12 months thereafter. Negative follow-up findings were determined by a 2-year cancer-free follow-up. Three AI computer-aided detection algorithms (AI-1, AI-2, and AI-3), sourced from different vendors, yielded a continuous score for the suspicion of cancer in each mammography examination. For a decision of normal or abnormal, the cut point was defined by the mean specificity of the first-reader radiologists (96.6%). RESULTSThe median age of study participants was 60 years (interquartile range, 50-66 years) for 739 women who received a diagnosis of breast cancer and 54 years (interquartile range, 47-63 years) for 8066 healthy controls. The cases positive for cancer comprised 618 (84%) screen detected and 121 (16%) clinically detected within 12 months of the screening examination. The area under the receiver operating curve for cancer detection was 0.956 (95% CI, 0.948-0.965) for AI-1, 0.922 (95% CI, 0.910-0.934) for AI-2, and 0.920 (95% CI, 0.909-0.931) for AI-3. At the specificity of the radiologists, the sensitivities were 81.9% for AI-1, 67.0% for AI-2, 67.4% for AI-3, 77.4% for first-reader radiologist, and 80.1% for second-reader radiologist. Combining AI-1 with first-reader radiologists achieved 88.6% sensitivity at 93.0% specificity (abnormal defined by either of the 2 making an abnormal assessment). No other examined combination of AI algorithms and radiologists surpassed this sensitivity level. CONCLUSIONS AND RELEVANCETo our knowledge, this study is the first independent evaluation of several AI computer-aided detection algorithms for screening mammography. The results of this study indicated that a commercially available AI computer-aided detection algorithm can assess screening mammograms with a sufficient diagnostic performance to be further evaluated as an independent reader in prospective clinical trials. Combining the first readers with the best algorithm identified more cases positive for cancer than combining the first readers with se...

AI-aided detection of malignant lesions in mammography screening – evaluation of a program in clinical practice

“… 3 Three commercially available AI algorithms have recently been subject to an external, independent evaluation in a case–control setting. 7 One of the studied algorithms showed a significantly higher performance compared with the other two algorithms, highlighting the importance of construction and training data for external validity. Interestingly, combining the best performing AI with the first reader/radiologist was more successful than combining the first and second readers regarding cancer detection rate, however, at the cost of increasing false-positive cases.…”

“…Interestingly, combining the best performing AI with the first reader/radiologist was more successful than combining the first and second readers regarding cancer detection rate, however, at the cost of increasing false-positive cases. 7 Another study, also evaluating a different AI program than Transpara, illustrates the complexity both in calibrating and in evaluating AI algorithms based on retrospective data. 8 For instance, reasonable cut-off levels must be chosen for the AI assessments (that are provided on a continuous scale) to enable comparison with the human assessments (that are provided on a binary scale: negative/suspected malignancy).…”

Studying human-AI collaboration protocols: the case of the Kasparov’s law in radiological double reading

“…Since the Kasparov’s Law (KL), as mentioned in the Introduction, can be framed as two different statements, we designed two different experiments: In order to evaluate the first KL statement, we used the InceptionV3 model as the “Strong” machine (which was actually more accurate than the best human reader involved in this study), and we then used the MobileNet model as the Machine (second reader) to be used in the “better” process of human-AI collaboration. Specifically, the Machine was always recruited as the second observer, like done in [ 39 ], where it was found that combining the first reader with the best algorithm identified more pathological cases than having a human as the second reader. In doing so, we considered a total of 90 3-reader 2-permutations.…”

Retrospective large-scale evaluation of an AI system as an independent reader for double reading in breast cancer screening

“…Recent works suggest that the current generation of AI-based algorithms may interpret mammograms at least at the level of human readers (19)(20)(21)(22)(23). Evaluations were based on small-scale reader studies (19)(20)(21) and larger scale retrospective evaluations (21)(22)(23) performed on artificially enriched sets, often involving resampling in an attempt to approximate a more representative screening population, and with dataset images significantly skewed towards a single mammography hardware vendor in each case. AI and its true potential to positively transform clinical practice on real-world screening populations remains to be confirmed.…”