Data Fusion Framework For The Prediction Of Early Hematoma Expansion Based On CNN

Spontaneous intracerebral hemorrhage (ICHs) is one of the common types of stroke. Patients with ICHs may bleed again in the early stage. The early hematoma expansion(HE) will increase the patient's mortality. Clinical status and laboratory parameters can be used as risk factors for HE. In terms of image signs, HE can be predicted based on the presence of independently predictable computed tomography(CT) signs. However, the sensitivity of these signs is not very high. Therefore, we propose an automatic framework of bimodal data fusion for the prediction of HE. The first pathway is a channel-attention-based encoder which extracts CT image features of brain hematoma and focuses on the lesion site. The second input is clinical parameters and this pathway only consists of transposed convolution. Finally, the outputs of two pathways are concatenated in the channel dimension. The experimental results indicate that our automatic framework based on CNN is superior to previous imaging-based methods, especially in terms of sensitivity. To the best of our knowledge, this is the first study that uses a fully automatic method to predict the early HE in patients with ICHs. Our framework achieved a comparable accuracy with the state-of-the-art method.