Multi-input segmentation of damaged brain in acute ischemic stroke patients using slow fusion with skip connection

Abstract

Time is a fundamental factor during stroke treatments. A fast, automatic approach that segments the ischemic regions helps treatment decisions. In clinical use today, a set of color-coded parametric maps generated from computed tomography perfusion (CTP) images are investigated manually to decide a treatment plan.
 We propose an automatic method based on a neural network using a set of parametric maps to segment the two ischemic regions (core and penumbra) in patients affected by acute ischemic stroke. Our model is based on a convolution-deconvolution bottleneck structure with multi-input and slow fusion. To address the data imbalance issue, a loss function based on the focal Tversky index is used. The proposed architecture demonstrates effective performance and results comparable to the ground truth annotated by neuroradiologists. A Dice coefficient of 0.81 for penumbra and 0.52 for core over the large vessel occlusion test set is achieved. The full implementation is available at: *************

Exploiting 4D CT Perfusion for segmenting infarcted areas in patients with suspected acute ischemic stroke

“…4), we implemented and compared the 2D-TCN [30] and the mJ-Net [32] to validate the performances of our models. Additionally, we compare the models with a method that uses a set of PMs as input [37]. In the remainder of the paper, we call this architecture Multi-input PMs.…”

“…Multi-input PMs [37] 3D-TCN-SE [33] Approach Figure 1: Visual comparison of the input for each implemented model. Every 4D CTP patient's study V ∈ R (X×Y ×Z×T ) undergoes a series of pre-processing steps to enhance each CTP scan (details in Algorithm 1).…”

“…Every 4D CTP patient's study V ∈ R (X×Y ×Z×T ) undergoes a series of pre-processing steps to enhance each CTP scan (details in Algorithm 1). Approach 1 (Multi-input PMs) and 2 (mJ-Net) are our methods implemented in [37] and [32], respectively. Approach 1 accepts a list of PMs generated from a CTP study in input instead of V .…”

“…An understanding of the extension of the penumbra during the first stages of the ischemic stroke is crucial for treatment decision [35,36]. To the best of our knowledge, Tomasetti et al was the first research group to segment both core and penumbra regions using machine learning [36] or DNN approaches [32,37].…”

“…In this paper, we present and investigate three novel models to segment the two ischemic regions (core and penumbra), where the input is the entire 4D CTP scans arranged in different ways to exploit the spatio-temporal nature of the data. We compare all models with previous work based on PMs [37] and slice-by-slice CTP [32], and two methods proposed by Amador et al [30,33].…”