Automated Detection of the Black Hole Sign for Patients with Intracerebral Hemorrhage Using Self-Supervised Learning

Intracerebral hemorrhage is a devastating form of stroke. Hematoma expansion (HE), growth of the hematoma on interval scans, predicts death and disability. Accurate prediction of HE is crucial for targeted interventions to improve patient outcomes. The black hole sign (BHS) on noncontrast CT scans is a predictive marker for HE. An automated method to recognize the BHS and predict HE could speed precise patient selection for treatment.

In this article, we present a novel framework leveraging self-supervised learning (SSL) techniques for BHS identification on head CT images. A ResNet-50 encoder model was pretrained on more than 1.7 million unlabeled head CT images. Layers for binary classification were added on top of the pretrained model. The resulting model was fine-tuned using the training data and evaluated on the held out test set to collect area under the curve (AUC) and F1 scores. The evaluations were performed on scan and slice levels. We ran different panels, one using 2 multicenter data sets for external validation and one including parts of them in the pretraining.

Our model demonstrated strong performance in identifying the BHS compared with the baseline model. Specifically, the model achieved scan-level AUC scores between 0.75 and 0.89 and F1 scores between 0.60 and 0.70. Furthermore, it exhibited robustness and generalizability across an external data set, achieving a scan-level AUC score of up to 0.85 and an F1 score of up to 0.60, while it performed worse on another data set with more heterogeneous samples. The negative effects could be mitigated after including parts of the external data sets in the fine-tuning process.

This study introduced a novel framework integrating SSL into medical image classification, particularly on BHS identification from head CT scans. The resulting pretrained head CT encoder model showed the potential to minimize manual annotation, which would significantly reduce labor, time, and costs. After fine-tuning, the framework demonstrated a promising performance for a specific downstream task, identifying the BHS to predict HE on comprehensive evaluation on diverse data sets. This approach holds promise for enhancing medical image analysis, particularly in scenarios with limited data availability.