Deep learning-based support systems have demonstrated encouraging results in numerous clinical applications involving the processing of time series data. While such systems often are very accurate, they have no inherent mechanism for explaining what influenced the predictions, which is critical for clinical tasks. However, existing explainability techniques lack an important component for trustworthy and reliable decision support, namely a notion of uncertainty. In this paper, we address this lack of uncertainty by proposing a deep ensemble approach where a collection of DNNs are trained independently. A measure of uncertainty in the relevance scores is computed by taking the standard deviation across the relevance scores produced by each model in the ensemble, which in turn is used to make the explanations more reliable. The class activation mapping method is used to assign a relevance score for each time step in the time series. Results demonstrate that the proposed ensemble is more accurate in locating relevant time steps and is more consistent across random initializations, thus making the model more trustworthy. The proposed methodology paves the way for constructing trustworthy and dependable support systems for processing clinical time series for healthcare related tasks.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To explore the standalone breast cancer detection performance at different risk score thresholds of a commercially available artificial intelligence (AI) system. Materials and Methods This retrospective study included information from 661,695 digital mammographic examinations performed among 242,629 female individuals screened as a part of x, 2004-2018. The study sample included 3807 screen-detected cancers (SDC) and 1110 interval breast cancers (IC). A continuous examination level risk score by the AI system was used to measure performance as the area under the receiver operating characteristic curve (AUC) with 95% CIs and cancer detection at different AI risk score thresholds. Results The AUC of the AI system was 0.93 (95% CI: 0.92-0.93) for SDC and IC combined and 0.97 (95% CI: 0.97-0.97) for SDC. In a setting where 10% of the examinations with the highest AI risk scores were defined as positive and 90% with the lowest scores as negative, 92.0% (3502/3807) of the SDC and 44.6% (495/1100) of the IC were identified by AI. In this scenario, 68.5% (10 987/16 029) of false positive screening results (negative recall assessment) were considered negative by AI. When 50% was used as the cut-off, 99.3% (3781/3807) of the SDC and 85.2% (946/1100) of the IC were identified as positive by AI, while 17.0% (2725/16 029) of the false positives were considered as negative. Conclusion The AI system showed high performance in detecting breast cancers within 2 years of screening mammography and a potential for triaging low-risk mammograms to reduce radiologist workload. ©RSNA, 2024.

Background Few studies have evaluated the role of artificial intelligence (AI) in prior screening mammography. Purpose To examine AI risk scores assigned to screening mammography in women who were later diagnosed with breast cancer. Materials and Methods Image data and screening information of examinations performed from January 2004 to December 2019 as part of BreastScreen Norway were used in this retrospective study. Prior screening examinations from women who were later diagnosed with cancer were assigned an AI risk score by a commercially available AI system (scores of 1-7, low risk of malignancy; 8-9, intermediate risk; and 10, high risk of malignancy). Mammographic features of the cancers based on the AI score were also assessed. The association between AI score and mammographic features was tested with a bivariate test. Results A total of 2787 prior screening examinations from 1602 women (mean age, 59 years ± 5.1 [SD]) with screen-detected (n = 1016) or interval (n = 586) cancers showed an AI risk score of 10 for 389 (38.3%) and 231 (39.4%) cancers, respectively, on the mammograms in the screening round prior to diagnosis. Among the screen-detected cancers with AI scores available two screening rounds (4 years) before diagnosis, 23.0% (122 of 531) had a score of 10. Mammographic features were associated with AI score for invasive screen-detected cancers (P < .001). Density with calcifications was registered for 13.6% (43 of 317) of screen-detected cases with a score of 10 and 4.6% (15 of 322) for those with a score of 1-7. Conclusion More than one in three cases of screen-detected and interval cancers had the highest AI risk score at prior screening, suggesting that the use of AI in mammography screening may lead to earlier detection of breast cancers. © RSNA, 2023 Supplemental material is available for this article. See also the editorial by Mehta in this issue.

Multivariate time series often contain missing values for reasons such as failures in data collection mechanisms. Since these missing values can complicate the analysis of time series data, imputation techniques are typically used to deal with this issue. However, the quality of the imputation directly affects the performance of downstream tasks. In this paper, we propose a selective imputation method that identifies a subset of timesteps with missing values to impute in a multivariate time series dataset. This selection, which will result in shorter and simpler time series, is based on both reducing the uncertainty of the imputations and representing the original time series as good as possible. In particular, the method uses multi-objective optimization techniques to select the optimal set of points, and in this selection process, we leverage the beneficial properties of the Multi-task Gaussian Process (MGP). The method is applied to different datasets to analyze the quality of the imputations and the performance obtained in downstream tasks, such as classification or anomaly detection. The results show that much shorter and simpler time series are able to maintain or even improve both the quality of the imputations and the performance of the downstream tasks.

Despite the significant improvements that self-supervised representation learning has led to when learning from unlabeled data, no methods have been developed that explain what influences the learned representation. We address this need through our proposed approach, RELAX, which is the first approach for attribution-based explanations of representations. Our approach can also model the uncertainty in its explanations, which is essential to produce trustworthy explanations. RELAX explains representations by measuring similarities in the representation space between an input and masked out versions of itself, providing intuitive explanations that significantly outperform the gradient-based baselines. We provide theoretical interpretations of RELAX and conduct a novel analysis of feature extractors trained using supervised and unsupervised learning, providing insights into different learning strategies. Moreover, we conduct a user study to assess how well the proposed approach aligns with human intuition and show that the proposed method outperforms the baselines in both the quantitative and human evaluation studies. Finally, we illustrate the usability of RELAX in several use cases and highlight that incorporating uncertainty can be essential for providing faithful explanations, taking a crucial step towards explaining representations.