MB-AI-His: Histopathological Diagnosis of Pediatric Medulloblastoma and its Subtypes via AI.

摘要: Medulloblastoma (MB) is a dangerous malignant pediatric brain tumor that could lead to death. It considered the most common cancerous tumor. Precise and timely diagnosis of MB its four subtypes (defined by World Health Organization (WHO)) essential decide appropriate follow-up plan suitable treatments prevent progression reduce mortality rates. Histopathology gold standard modality for subtypes, but manual via pathologist very complicated, needs excessive time, subjective pathologists’ expertise skills, which may variability in or misdiagnosis. The main purpose paper propose time-efficient reliable computer-aided (CADx), namely MB-AI-His, automatic from histopathological images. challenge this work lack datasets available limited related work. Related studies are based on either textural analysis deep learning (DL) feature extraction methods. These used individual features perform classification task. However, MB-AI-His combines benefits DL techniques methods through cascaded manner. First, it uses three convolutional neural networks (CNNs), including DenseNet-201, MobileNet, ResNet-50 CNNs extract spatial features. Next, extracts time-frequency discrete wavelet transform (DWT), method. Finally, fuses spatial-time-frequency generated DWT using cosine (DCT) principal component (PCA) produce CADx system. merges privileges different CNN architectures. has binary level classifying among normal abnormal images, multi-classification classify MB. results show accurate both multi-class levels. also system as PCA DCT have efficiently reduced training execution time. performance compared with systems, comparison verified powerfulness outperforming results. Therefore, can support pathologists time cost procedure will correspondingly lower death