With an increase in deep learning-based methods, the call for explainability of such methods grows, especially in high-stakes decision making areas such as medical image analysis. This survey presents an overview of explainable artificial intelligence (XAI) used in deep learning-based medical image analysis. A framework of XAI criteria is introduced to classify deep learning-based medical image analysis methods. Papers on XAI techniques in medical image analysis are then surveyed and categorized according to the framework and according to anatomical location. The paper concludes with an outlook of future opportunities for XAI in medical image analysis.

To retrospectively investigate whether parenchymal enhancement in dynamic contrast material-enhanced magnetic resonance (MR) imaging of the contralateral breast in patients with unilateral invasive breast cancer is associated with therapy outcome. After obtaining approval of the institutional review board and patients' written informed consent, 531 women with unilateral invasive breast cancer underwent dynamic contrast-enhanced MR imaging between 2000 and 2008. The contralateral parenchyma was segmented automatically, in which the mean of the top 10% late enhancement was calculated. Cox regression was used to test associations between parenchymal enhancement, patient and tumor characteristics, and overall survival and invasive disease-free survival. Subset analyses were performed and stratified according to immunohistochemical subtypes and type of adjuvant treatment received. Median follow-up was 86 months. Age (P < .001) and immunohistochemical subtype (P = .042) retained significance in multivariate analysis for overall survival. In patients with estrogen receptor-positive and human epidermal growth factor receptor 2 (HER2)-negative breast cancer (n = 398), age (P < .001), largest diameter on MR images (P = .049), and parenchymal enhancement (P = .011) were significant. In patients who underwent endocrine therapy (n = 174), parenchymal enhancement was the only significant covariate for overall survival and invasive disease-free survival (P < .001). Results suggest that parenchymal enhancement in the contralateral breast of patients with invasive unilateral breast cancer is significantly associated with long-term outcome, particularly in patients with estrogen receptor-positive, human epidermal growth factor receptor 2-negative breast cancer. Lower value of the mean top 10% enhancement of the parenchyma shows potential as a predictive biomarker for relatively poor outcome in patients who undergo endocrine therapy. These results should, however, be validated in a larger study.

In women with extremely dense breasts, a deep learning model identified breast MRI examinations without cancer with high certainty, dismissing nearly 40% of normal MRI examinations without missing ... Background Supplemental screening with MRI has proved beneficial in women with extremely dense breasts. Most MRI examinations show normal anatomic and physiologic variation that may not require rad...

To purpose of this paper was to assess the feasibility of volumetric breast density estimations on MRI without segmentations accompanied with an explainability step. A total of 615 patients with breast cancer were included for volumetric breast density estimation. A 3-dimensional regression convolutional neural network (CNN) was used to estimate the volumetric breast density. Patients were split in training (N = 400), validation (N = 50), and hold-out test set (N = 165). Hyperparameters were optimized using Neural Network Intelligence and augmentations consisted of translations and rotations. The estimated densities were evaluated to the ground truth using Spearman’s correlation and Bland–Altman plots. The output of the CNN was visually analyzed using SHapley Additive exPlanations (SHAP). Spearman’s correlation between estimated and ground truth density was ρ = 0.81 (N = 165, P < 0.001) in the hold-out test set. The estimated density had a median bias of 0.70% (95% limits of agreement = − 6.8% to 5.0%) to the ground truth. SHAP showed that in correct density estimations, the algorithm based its decision on fibroglandular and fatty tissue. In incorrect estimations, other structures such as the pectoral muscle or the heart were included. To conclude, it is feasible to automatically estimate volumetric breast density on MRI without segmentations, and to provide accompanying explanations.

MRI-derived radiomics features were associated with protein expression related to the ribosome (targeted in anticancer drugs) and genes related to the extracellular matrix (involved in cancer progr... Background Better understanding of the molecular biology associated with MRI phenotypes may aid in the diagnosis and treatment of breast cancer. Purpose To discover the associations between MRI phe...