MICCAI 2021 Daily - Tuesday

Two large datasets that Third Eye uses are the MIMIC dataset from MIT , one of the richest ICU databases, and another large collaborative research database called eICU . These include data of thousands of patients recorded for over three years. This includes not only vital signs in the form of signals and waveforms, but also raw patient notes, such as radiology reports, discharge notes, and patient consultation notes. Third Eye did not want to limit themselves to using just one source of data, to avoid learning internal biases or a situation where a data source becomes unavailable. “ We decided to create a multimodal deep learning system that would learn joint features from both text and signals, ” Samyakh explains. “ We trained by a representation learning . Instead of predicting organ failure, we try to represent what the patient is going through as well as possible in relation to other patients and their trajectories. Then we can infer indirectly when the 21 DAILY MICCAI Tuesday Third Eye Intelligence Resulting organ failure prediction model developed by the AutoML algorithm