RoMIA: a framework for creating Robust Medical Imaging AI models for chest radiographs.

Artificial Intelligence (AI) methods, particularly Deep Neural Networks (DNNs), have shown great promise in a range of medical imaging tasks. However, the susceptibility of DNNs to producing erroneous outputs under the presence of input noise and variations is of great concern and one of the largest challenges to their adoption in medical settings. Towards addressing this challenge, we explore the robustness of DNNs trained for chest radiograph classification under a range of perturbations reflective of clinical settings. We propose RoMIA, a framework for the creation of Robust Medical Imaging AI models. RoMIA adds three key steps to the model training and deployment flow: (i) Noise-added training, wherein a part of the training data is synthetically transformed to represent common noise sources, (ii) Fine-tuning with input mixing, in which the model is refined with inputs formed by mixing data from the original training set with a small number of images from a different source, and (iii) DCT-based denoising, which removes a fraction of high-frequency components of each image before applying the model to classify it. We applied RoMIA to create six different robust models for classifying chest radiographs using the CheXpert dataset. We evaluated the models on the CheXphoto dataset, which consists of naturally and synthetically perturbed images intended to evaluate robustness. Models produced by RoMIA show 3%-5% improvement in robust accuracy, which corresponds to an average reduction of 22.6% in misclassifications. These results suggest that RoMIA can be a useful step towards enabling the adoption of AI models in medical imaging applications.

Time-varying model of engagement with digital self reporting: Evidence from smoking cessation longitudinal studies.

Objective: Insufficient engagement is a critical barrier impacting the utility of digital interventions and mobile health assessments. As a result, engagement itself is increasingly becoming a target of studies and interventions. The purpose of this study is to investigate the dynamics of engagement in mobile health data collection by exploring whether, how, and why response to digital self-report prompts change over time in smoking cessation studies. Method: Data from two ecological momentary assessment (EMA) studies of smoking cessation among diverse smokers attempting to quit (N = 573) with a total of 65,974 digital self-report prompts. We operationalize engagement with self-reporting in term of prompts delivered and prompt response to capture both broad and more granular engagement in self-reporting, respectively. The data were analyzed to describe trends in prompt delivered and prompt response over time. Time-varying effect modeling (TVEM) was employed to investigate the time-varying effects of response to previous prompt and the average response rate on the likelihood of current prompt response. Results: Although prompt response rates were relatively stable over days in both studies, the proportion of participants with prompts delivered declined steadily over time in one of the studies, indicating that over time, fewer participants charged the device and kept it turned on (necessary to receive at least one prompt per day). Among those who did receive prompts, response rates were relatively stable. In both studies, there is a significant, positive and stable relationship between response to previous prompt and the likelihood of response to current prompt throughout all days of the study. The relationship between the average response rate prior to current prompt and the likelihood of responding to the current prompt was also positive, and increasing with time. Conclusion: Our study highlights the importance of integrating various indicators to measure engagement in digital self-reporting. Both average response rate and response to previous prompt were highly predictive of response to the next prompt across days in the study. Dynamic patterns of engagement in digital self-reporting can inform the design of new strategies to promote and optimize engagement in digital interventions and mobile health studies.

Quantifying the Brain Predictivity of Artificial Neural Networks With Nonlinear Response Mapping.

Quantifying the similarity between artificial neural networks (ANNs) and their biological counterparts is an important step toward building more brain-like artificial intelligence systems. Recent efforts in this direction use neural predictivity, or the ability to predict the responses of a biological brain given the information in an ANN (such as its internal activations), when both are presented with the same stimulus. We propose a new approach to quantifying neural predictivity by explicitly mapping the activations of an ANN to brain responses with a non-linear function, and measuring the error between the predicted and actual brain responses. Further, we propose to use a neural network to approximate this mapping function by training it on a set of neural recordings. The proposed method was implemented within the TensorFlow framework and evaluated on a suite of 8 state-of-the-art image recognition ANNs. Our experiments suggest that the use of a non-linear mapping function leads to higher neural predictivity. Our findings also reaffirm the observation that the latest advances in classification performance of image recognition ANNs are not matched by improvements in their neural predictivity. Finally, we examine the impact of pruning, a widely used ANN optimization, on neural predictivity, and demonstrate that network sparsity leads to higher neural predictivity.

Detection of correlated dynamics on multiple timescales by measurement of the differential relaxation of zero- and double-quantum coherences involving sidechain methyl groups in proteins.

Multiple effects may lead to significant differences between the relaxation rates of zero-quantum coherences (ZQC) and double-quantum coherences (DQC) generated between a pair of nuclei in solution. These include the interference between the anisotropic chemical shifts of the two nuclei participating in formation of the ZQC or DQC, the individual dipolar interactions of each of the two nuclei with the same proton, and the slow modulation of the isotropic chemical shifts of the two nuclei due to conformational exchange. Motional events that occur on a timescale much faster than the rotational correlation time (ps-ns) influence the first two effects, while the third results from processes that occur on a far slower timescale (mus-ms). An analysis of the differential relaxation of ZQC and DQC is thus informative about dynamics on the fast as well as the slow timescales. We present here an experiment that probes the differential relaxation of ZQC and DQC involving methyl groups in protein sidechains as an extension to our recently proposed experiments for the protein backbone. We have applied the methodology to (15)N, (13)C-labeled ubiquitin and used a detailed analysis of the measured relaxation rates using a simple single-axis diffusion model to probe the motional restriction of the C(next)H(next) bond vector where C(next) is the carbon that is directly bonded to a sidechain methyl carbon (C(methyl)). Comparison of the present results with the motional restriction of the C(next)C(methyl) bond (S(axis)(2)) reveals that the single-axis diffusion model, while valid in the fringes of the protein and for shorter chain amino acids, proves inadequate in the central protein core for long chain, asymmetrically branched amino acids where more complex motional models are necessary, as is the inclusion of the possibility of correlation between multiple motional modes. In addition, the present measurements report on the modulation of isotropic chemical shifts due to motion on the mus-ms timescale. Three Leu residues (8, 50, and 56) are found to display these effects. These residues lie in regions where chemical shift modulation had been detected previously both in the backbone and sidechain regions of ubiquitin.