Detection of Pleuropulmonary Blastoma at an Early Stage Using Vision Transformer Model

Abstract

Childhood cancer is the second most common cause of death in children under the age of fifteen, according to the American Cancer Society, and the incidence of diagnosis is rising. One common cancer is pleuropulmonary blastoma (PPB), which affects newborns to six-year-old children. Clinical diagnosis is through imaging, which is speedy and economical and does not require specialized equipment or laboratory tests. Still, it can be challenging to analyze PPB early using only imaging, and identifying clinical signs may also pose a challenge due to the numerous possible differential diagnoses. Clinical methods are unreliable for fast and accurate results, time-consuming, and prone to errors. Detecting PPB at an early stage is essential for its proper treatment, as it can be fatal if left untreated. In the last few years, convolutional neural networks (CNNs) have become the most prevalent technique for computer vision tasks. However, CNNs have a restricted local receptive field that may hinder their ability to learn about the global context. An alternative approach to CNNs that looks promising is the Vision Transformer (ViT). ViT utilizes self-attention between image patches to process visual information. This experiment uses the ViT base Model, an advanced deep-learning algorithm, to overcome these difficulties. ViT not only reduces the computation but also achieves better results than CNN. Our experiments with (LIDC-IDRI), include different models of medical imaging, such as CT, DX, and CR, and consist of 244,527 images. The proposed model evaluates the cancerous cells in the histopathological images to determine and detect PPB disease. From the result of the experiment, the efficiency of the proposed ViT model is verified and compared with other traditional clinical models and the DCNN model to evaluate the performance. The outcome shows that the accuracy and sensitivity of the method proposed in this research reach 99.47% and 99.9% for the medical imaging dataset.

Keywords

• pleuropulmonary blastoma (ppb)
• lung cancer
• cancer detection
• diagnosis
• medical imaging
• histopathological imaging
• vision transformer (vit).