Sessions

The MIE conference has always been a hub for innovation and cutting-edge technology. This year was no exception several demonstrators showcase their unique solutions to healthcare problems. Use the opportunity to take a closer look at some of the most striking demonstrators present at the conference, highlighting their diversity and the potential impact they could have on various healthcare-related challenges

Deep learning models for radiology are typically deployed either through cloud-based platforms, through on-premises infrastructures, or though heavyweight viewers. This tends to restrict the audience of deep learning models to radiologists working in state-of-the-art hospitals, which raises concerns about the democratization of deep learning for medical imaging, most notably in the context of research and education. We show that complex deep learning models can be applied directly inside Web browsers, without resorting to any external computation infrastructure, thanks to the use of WebAssembly, and we release our code as free and open-source software. This opens the path to the use of teleradiology solutions as an effective way to distribute, teach, and evaluate deep learning architectures.

The automatic de-identification of clinical narrative text offers efficient patient data privacy protection and eases reuse of clinical data. CliniDeID applies an ensemble method combining deep and shallow machine learning with rule-based algorithms and is released as free and open-source solution to de-identify unstructured clinical text with high accuracy. It was recently evaluated with a selection of clinical text corpora and reached high sensitivity and positive predictive value.

We present a study comparing two mediums that can be used to communicate with allophone patients: a speech-enabled phraselator (BabelDr) and telephone interpreting. 

Electrodermal activity (EDA) reflects sympathetic nervous system activitythrough sweating-related changes in skin conductance. Decomposition analysisis used to deconvolve the EDA into slow and fast varying tonic and phasicactivity, respectively. In this study, we used machine learning models to comparethe performance of two EDA decomposition algorithms to detect emotions such asamusing, boring, relaxing, and scary. The EDA data considered in this study wereobtained from the publicly available Continuously Annotated Signals of Emotion(CASE) dataset. Initially, we pre-processed and deconvolved the EDA data intotonic and phasic components using decomposition methods such as cvxEDA andBayesianEDA. Further, twelve-time domain features were extracted from the phasiccomponent of EDA data. Finally, we applied machine learning techniques suchas logistic regression (LR) and support vector machine (SVM), to evaluate the performanceof the decomposition method. Our results imply that the BayesianEDAdecomposition method outperforms the cvxEDA. The mean of first derivative featurediscriminated all the considered emotional pairs with high statistical significance(p<0.05). SVM was able to detect emotions better than the LR classifier.We achieved a 10-fold average classification accuracy, sensitivity, specificity, precision,and f1-score of 88.2%, 76.25%, 92.08%, 76.16%, and 76.15% respectively,using BayesianEDA and SVM classifier. The proposed framework can be utilizedto detect emotional states for the early diagnosis of psychological conditions.