Evaluation of Combined Artificial Intelligence and Radiologist Assessment to Interpret Screening Mammograms.

Importance: Mammography screening currently relies on subjective human interpretation. Artificial intelligence (AI) advances could be used to increase mammography screening accuracy by reducing missed cancers and false positives. Objective: To evaluate whether AI can overcome human mammography interpretation limitations with a rigorous, unbiased evaluation of machine learning algorithms. Design, Setting, and Participants: In this diagnostic accuracy study conducted between September 2016 and November 2017, an international, crowdsourced challenge was hosted to foster AI algorithm development focused on interpreting screening mammography. More than 1100 participants comprising 126 teams from 44 countries participated. Analysis began November 18, 2016. Main Outcomes and Measurements: Algorithms used images alone (challenge 1) or combined images, previous examinations (if available), and clinical and demographic risk factor data (challenge 2) and output a score that translated to cancer yes/no within 12 months. Algorithm accuracy for breast cancer detection was evaluated using area under the curve and algorithm specificity compared with radiologists' specificity with radiologists' sensitivity set at 85.9% (United States) and 83.9% (Sweden). An ensemble method aggregating top-performing AI algorithms and radiologists' recall assessment was developed and evaluated. Results: Overall, 144 231 screening mammograms from 85 580 US women (952 cancer positive ≤12 months from screening) were used for algorithm training and validation. A second independent validation cohort included 166 578 examinations from 68 008 Swedish women (780 cancer positive). The top-performing algorithm achieved an area under the curve of 0.858 (United States) and 0.903 (Sweden) and 66.2% (United States) and 81.2% (Sweden) specificity at the radiologists' sensitivity, lower than community-practice radiologists' specificity of 90.5% (United States) and 98.5% (Sweden). Combining top-performing algorithms and US radiologist assessments resulted in a higher area under the curve of 0.942 and achieved a significantly improved specificity (92.0%) at the same sensitivity. Conclusions and Relevance: While no single AI algorithm outperformed radiologists, an ensemble of AI algorithms combined with radiologist assessment in a single-reader screening environment improved overall accuracy. This study underscores the potential of using machine learning methods for enhancing mammography screening interpretation.

Automatic intensity windowing of mammographic images based on a perceptual metric.

PURPOSE: Initial auto-adjustment of the window level WL and width WW applied to mammographic images. The proposed intensity windowing (IW) method is based on the maximization of the mutual information (MI) between a perceptual decomposition of the original 12-bit sources and their screen displayed 8-bit version. Besides zoom, color inversion and panning operations, IW is the most commonly performed task in daily screening and has a direct impact on diagnosis and the time involved in the process. METHODS: The authors present a human visual system and perception-based algorithm named GRAIL (Gabor-relying adjustment of image levels). GRAIL initially measures a mammogram's quality based on the MI between the original instance and its Gabor-filtered derivations. From this point on, the algorithm performs an automatic intensity windowing process that outputs the WL/WW that best displays each mammogram for screening. GRAIL starts with the default, high contrast, wide dynamic range 12-bit data, and then maximizes the graphical information presented in ordinary 8-bit displays. Tests have been carried out with several mammogram databases. They comprise correlations and an ANOVA analysis with the manual IW levels established by a group of radiologists. A complete MATLAB implementation of GRAIL is available at https://github.com/TheAnswerIsFortyTwo/GRAIL. RESULTS: Auto-leveled images show superior quality both perceptually and objectively compared to their full intensity range and compared to the application of other common methods like global contrast stretching (GCS). The correlations between the human determined intensity values and the ones estimated by our method surpass that of GCS. The ANOVA analysis with the upper intensity thresholds also reveals a similar outcome. GRAIL has also proven to specially perform better with images that contain micro-calcifications and/or foreign X-ray-opaque elements and with healthy BI-RADS A-type mammograms. It can also speed up the initial screening time by a mean of 4.5 s per image. CONCLUSIONS: A novel methodology is introduced that enables a quality-driven balancing of the WL/WW of mammographic images. This correction seeks the representation that maximizes the amount of graphical information contained in each image. The presented technique can contribute to the diagnosis and the overall efficiency of the breast screening session by suggesting, at the beginning, an optimal and customized windowing setting for each mammogram.