Detecting abnormal thyroid cartilages on CT using deep learning.

PURPOSE: The purpose of this study was to evaluate the performance of a deep learning algorithm in detecting abnormalities of thyroid cartilage from computed tomography (CT) examination. MATERIALS AND METHODS: A database of 515 harmonized thyroid CT examinations was used, of which information regarding cartilage abnormality was provided for 326. The process consisted of determining image abnormality and, from these preprocessed images, finding the best learning algorithm to appropriately characterize thyroid cartilage as normal or abnormal. CT images were cropped to be centered around the cartilage in order to focus on the relevant area. New images were generated from the originals by applying simple transformations in order to augment the database. Characterizations of cartilage abnormalities were made using transfer learning, by using the architecture of a pre-trained neural network called VGG16 and adapting the final layers to a binary classification problem. RESULTS: The best algorithm yielded an area under the receiving operator characteristic curve (AUC) of 0.72 on a sample of 82 thyroid test images. The sensitivity and specificity of the abnormality detection were 83% and 64% at the best threshold, respectively. Applying the model on another independent sample of 189 new thyroid images resulted in an AUC of 0.70. CONCLUSION: This study demonstrates the feasibility of using a deep learning-based abnormality detection system to evaluate thyroid cartilage from CT examinations. However, although promising results, the model is not yet able to match an expert's diagnosis.

Single-trial decoding of auditory novelty responses facilitates the detection of residual consciousness.

Detecting residual consciousness in unresponsive patients is a major clinical concern and a challenge for theoretical neuroscience. To tackle this issue, we recently designed a paradigm that dissociates two electro-encephalographic (EEG) responses to auditory novelty. Whereas a local change in pitch automatically elicits a mismatch negativity (MMN), a change in global sound sequence leads to a late P300b response. The latter component is thought to be present only when subjects consciously perceive the global novelty. Unfortunately, it can be difficult to detect because individual variability is high, especially in clinical recordings. Here, we show that multivariate pattern classifiers can extract subject-specific EEG patterns and predict single-trial local or global novelty responses. We first validate our method with 38 high-density EEG, MEG and intracranial EEG recordings. We empirically demonstrate that our approach circumvents the issues associated with multiple comparisons and individual variability while improving the statistics. Moreover, we confirm in control subjects that local responses are robust to distraction whereas global responses depend on attention. We then investigate 104 vegetative state (VS), minimally conscious state (MCS) and conscious state (CS) patients recorded with high-density EEG. For the local response, the proportion of significant decoding scores (M=60%) does not vary with the state of consciousness. By contrast, for the global response, only 14% of the VS patients' EEG recordings presented a significant effect, compared to 31% in MCS patients' and 52% in CS patients'. In conclusion, single-trial multivariate decoding of novelty responses provides valuable information in non-communicating patients and paves the way towards real-time monitoring of the state of consciousness.