Machine learning (ML) models have been shown to predict the presence of clinical factors from medical imaging with remarkable accuracy. However, these complex models can be difficult to interpret and are often criticized as “black boxes”. Prediction models that provide no insight into how their predictions are obtained are difficult to trust for making important clinical decisions, such as medical diagnoses or treatment. Explainable machine learning (XML) methods, such as Shapley values, have made it possible to explain the behavior of ML algorithms and to identify which predictors contribute most to a prediction. Incorporating XML methods into medical software tools has the potential to increase trust in ML-powered predictions and aid physicians in making medical decisions. Specifically, in the field of medical imaging analysis the most used methods for explaining deep learning-based model predictions are saliency maps that highlight important areas of an image. However, they do not provide a straightforward interpretation of which qualities of an image area are important. Here, we describe a novel pipeline for XML imaging that uses radiomics data and Shapley values as tools to explain outcome predictions from complex prediction models built with medical imaging with well-defined predictors. We present a visualization of XML imaging results in a clinician-focused dashboard that can be generalized to various settings. We demonstrate the use of this workflow for developing and explaining a prediction model using MRI data from glioma patients to predict a genetic mutation.

Diagnosis of adamantinomatous craniopharyngioma (ACP) is predominantly determined through invasive pathological examination of a neurosurgical biopsy specimen. Clinical experts can distinguish ACP from Magnetic Resonance Imaging (MRI) with an accuracy of 86%, and 9% of ACP cases are diagnosed this way. Classification using deep learning (DL) provides a solution to support a non-invasive diagnosis of ACP through neuroimaging, but it is still limited in implementation, a major reason being the lack of predictive uncertainty representation. We trained and tested a DL classifier on preoperative MRI from 86 suprasellar tumor patients across multiple institutions. We then applied a Bayesian DL approach to calibrate our previously published ACP classifier, extending beyond point-estimate predictions to predictive distributions. Our original classifier outperforms random forest and XGBoost models in classifying ACP. The calibrated classifier underperformed our previously published results, indicating that the original model was overfit. Mean values of the predictive distributions were not informative regarding model uncertainty. However, the variance of predictive distributions was indicative of predictive uncertainty. We developed an algorithm to incorporate predicted values and the associated uncertainty to create a classification abstention mechanism. Our model accuracy improved from 80.8% to 95.5%, with a 34.2% abstention rate. We demonstrated that calibration of DL models can be used to estimate predictive uncertainty, which may enable clinical translation of artificial intelligence to support non-invasive diagnosis of brain tumors in the future.

HIV type 1 is associated with pulmonary dysfunction that is exacerbated by cigarette smoke. Alveolar macrophages (AM) are the most prominent immune cell in the alveolar space. These cells play an important role in clearing inhaled pathogens and regulating the inflammatory environment; however, how HIV infection impacts AM phenotype and function is not well understood, in part because of their autofluorescence and the absence of well-defined surface markers. The main aim of this study was to evaluate the impact of HIV infection on human AM and to compare the effect of smoking on their phenotype and function. Time-of-flight mass cytometry and RNA sequencing were used to characterize macrophages from human bronchoalveolar lavage of HIV-infected and -uninfected smokers and nonsmokers. We found that the frequency of CD163+ anti-inflammatory AM was decreased, whereas CD163−CCR7+ proinflammatory AM were increased in HIV infection. HIV-mediated proinflammatory polarization was associated with increased levels of inflammatory cytokines and macrophage activation. Conversely, smoking heightened the inflammatory response evident by change in the expression of CXCR4 and TLR4. Altogether, these findings suggest that HIV infection, along with cigarette smoke, favors a proinflammatory macrophage phenotype associated with enhanced expression of inflammatory molecules. Further, this study highlights time-of-flight mass cytometry as a reliable method for immunophenotyping the highly autofluorescent cells present in the bronchoalveolar lavage of cigarette smokers.

Network visualizations are understudied in graphical perception. As a result, most network visualization designs still largely rely on designer intuition and algorithm optimizations rather than being guided by knowledge of human perception. The lack of perceptual understanding of network visualizations also limits the generalizability of past empirical evaluations, given their focus on performance over causal interpretation. To bridge this gap between perception and network visualization, we introduce a framework highlighting five key perceptual mechanisms used in node-link diagrams and adjacency matrices: attention, visual search, perceptual organization, ensemble coding, and object recognition. Our framework describes the role these perceptual mechanisms play in common network analytical tasks. We use the framework to revisit four past empirical investigations and outline future design experiments that can help produce more perceptually effective network visualizations. We anticipate this connection will afford translational understanding to guide more effective network visualization design and offer hypotheses for perception-aware network visualizations.

Artificial Intelligence (AI) is well-suited to help support complex decision-making tasks within clinical medicine, including clinical imaging applications like radiographic differential diagnosis of central nervous system (CNS) tumors. So far, there have been numerous examples of theoretical AI solutions for this space, for example, large-scale corporate efforts like IBM’s Watson AI. However, clinical implementation remains limited due to factors related to the alignment of this technology in the clinical setting. User-Centered Design (UCD) is a design philosophy that focuses on developing tailored solutions for specific users or user groups. In this study, we applied UCD to develop an explainable AI tool to support clinicians in our use case. Through four design iterations, starting from basic functionality and visualizations, we progressed to functional prototypes in a realistic testing environment. We discuss our motivation and approach for each iteration, along with key insights gained. This UCD process has advanced our conceptual idea from feasibility testing to interactive functional AI interfaces designed for specific clinical and cognitive tasks. It has also provided us with directions to develop further an AI system for the non-invasive diagnosis of CNS tumors.