Deep Sequential Learning For Cervical Spine Fracture Detection On Computed Tomography Imaging

Abstract

Fractures of the cervical spine are a medical emergency and may lead to permanent paralysis and even death. Accurate diagnosis in patients with suspected fractures by computed tomography (CT) is critical to patient management. In this paper, we propose a deep convolutional neural network (DCNN) with a bidirectional long-short term memory (BLSTM) layer for the automated detection of cervical spine fractures in CT axial images. We used an annotated dataset of 3,666 CT scans (729 positive and 2,937 negative cases) to train and validate the model. The validation results show a classification accuracy of 70.92% and 79.18% on the balanced (104 positive and 104 negative cases) and imbalanced (104 positive and 419 negative cases) test datasets, respectively.

Computer Aided Facial Bone Fracture Diagnosis (CA-FBFD) System Based on Object Detection Model

“…Hojjat Salehinejad et al [25] raised the problem of disproportionate fracture data. Compared to the amount of data for normal persons without fractures, that for patients with fractures is extremely small.…”

“…Furthermore, even in the CT images of patients with fractures, most CT images of patients with fractures do not contain fracture parts because they occupy only a tiny area on a face. Hojjat Salehinejad et al [25] proposed a bidirectional long and short-term memory model to address these issues. Amelia Jim´enez-S´anchez et al [24] investigated the classification of femur fractures into seven types based on X-ray images, with six types classified based on fracture location and, type, and the number of fragments in the fracture, and the remaining type classified as a normal class without fracture.…”

“…Curriculum learning is a method of scheduling key training samples to learn more by scoring training samples. Hojjat Salehinejad et al [25] can determine an approximate location of the fracture using the Grad-CAM [33], and Amelia Jim´enez-S´anchez et al [24] can determine whether the fracture is anterior or posterior through location classification. However, both algorithms have limitations in their ability to express the location of a fracture.…”

“…Imbalanced datasets cause serious problems in classification tasks in machine learning (ML) [32]. As Hojjat Salehinejad et al [25] points out, there is a data imbalance problem because there are significantly fewer fracture data than non-fracture data. The multi-positive strategy is a sampling strategy that trains a deep learning model by considering the predicted bounding box in a 3x3 region centered on the fracture location as positive samples, which is called "center sampling" in FCOS [15], which alleviates the problem of data imbalance in fracture data by compensating for the insufficient number of positive samples.…”

S-Rocket: Selective Random Convolution Kernels for Time Series Classification

“…Mostly, time series refers to a regular time series in the literature [1]. Time series occurs in various real-world applications such as human activity recognition [2], travel mode detection [3], natural language processing [4], speech recognition [5], shopping pattern recognition [6], electronic health records [7], and medical imaging [8].…”