Abstract
To analyze the brain CT imaging data of children with cerebral palsy (CP), deep learning-based electronic computed tomography (CT) imaging information characteristics were used, thereby providing help for the rehabilitation analysis of children with CP and comorbid epilepsy. The brain CT imaging data of 73 children with CP were collected, who were outpatients or inpatients in our hospital. The images were randomly divided into two groups. One group was the artificial intelligence image group, and hybrid segmentation network (HSN) model was employed to analyze brain images to help the treatment. The other group was the control group, and original images were used to help diagnosis and treatment. The deep learning-based HSN was used to segment the CT image of the head of patients and was compared with other CNN methods. It was found that HSN had the highest Dice score (DSC) among all models. After treatment, six cases in the artificial intelligence image group returned to normal (20.7%), and the artificial intelligence image group was significantly higher than the control group (X2 = 335191, P < 0.001). The cerebral hemodynamic changes were obviously different in the two groups of children before and after treatment. The VP of the cerebral artery in the child was (139.68 ± 15.66) cm/s after treatment, which was significantly faster than (131.84 ± 15.93) cm/s before treatment, P < 0.05. To sum up, the deep learning model can effectively segment the CP area, which can measure and assist the diagnosis of future clinical cases of children with CP. It can also improve medical efficiency and accurately identify the patient's focus area, which had great application potential in helping to identify the rehabilitation training results of children with CP.
Highlights
Cerebral palsy (CP) is one of the common causes of disability in children
Neuroimaging examination is an important auxiliary examination for central nervous system damage, which can provide objective basis for changes in tissue morphology for clinical diagnosis and treatment. e traditional cranial computed tomography (CT) has been widely used in the cranial imaging examination of children with CP and has accumulated certain experience [4]. e other functional imaging examinations developed based on traditional Journal of Healthcare Engineering techniques such as magnetic resonance imaging (MRI), ultrasound (US), positron emission tomography (PET), and other devices can assess the function of brain tissue through local blood flow changes, water molecular activity, and metabolic status. e lesions associated with the occurrence of CP were mapped in more detail to provide evidence of functional and metabolic abnormalities for some lesions with insignificant morphological changes
Comparative Test with 2DCNN Method. It was compared with 2D CNN method. e 2D models allowed larger images as input than 3D models (Figure 2). erefore, full-resolution CT slices were used to collect detailed contextual information. e 2D CNN similar to 3D CNN was constructed, and the difference was that 3D convolution was replaced with 2D convolution. e 3D trilinear upsampling layer was replaced with 2D bilinear upsampling layer, and batch normalization was used in convolution
Summary
Cerebral palsy (CP) is one of the common causes of disability in children. Neuroimaging examination is an important auxiliary examination for central nervous system damage, which can provide objective basis for changes in tissue morphology for clinical diagnosis and treatment. Journal of Healthcare Engineering techniques such as magnetic resonance imaging (MRI), ultrasound (US), positron emission tomography (PET), and other devices can assess the function of brain tissue through local blood flow changes, water molecular activity, and metabolic status. In some cases, general radiologists must make a diagnosis every three to four seconds in an 8-hour working day to meet the needs of the workload [5]. Some diseases do not cause significant structural changes and require high resolution and small-scale brain imaging techniques. The patient may suffer from ten related diseases. (III) e excessive number of extracted features consumes a large amount of storage space and at the same time greatly increases the computational complexity and leads to dimensional disaster. (IV) e classification accuracy should be further improved to meet the requirements of practical use. (V) e generalization performance of the classifier is poor, and the prediction effect of the new samples is obviously lower than that of the training samples
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
