FibroVit—Vision transformer-based framework for detection and classification of pulmonary fibrosis from chest CT images

• DOI: 10.3389/fmed.2023.1282200
• Publication Date: 2023-11-09T16:17:20Z
• Authors: Muhammad Waseem S...; Muhammad Farhan; Nabil Sharaf Almalki; Mrim M. Alnfiai; Gabriel Avelino S...
• Keywords: Hyperparameter

Pulmonary Fibrosis (PF) is an immedicable respiratory condition distinguished by permanent fibrotic alterations in the pulmonary tissue for which there no cure. Hence, it crucial to diagnose PF swiftly and precisely. The existing research on deep learning-based fibrosis detection methods has limitations, including dataset sample sizes a lack of standardization data preprocessing evaluation metrics. This study presents comparative analysis four vision transformers regarding their efficacy accurately detecting classifying patients with ability localize abnormalities within Images obtained from Computerized Tomography (CT) scans. consisted 13,486 samples selected out 24647 dataset, included both PF-positive CT normal images that underwent preprocessing. preprocessed were divided into three sets: training set, accounted 80% total pictures; validation comprised 10%; test also 10%. transformer models, ViT, MobileViT2, ViTMSN, BEiT subjected procedures, during hyperparameters like learning rate batch size fine-tuned. overall performance optimized architectures been assessed using various metrics showcase consistent fine-tuned model. Regarding performance, ViT shown superior testing accuracy loss minimization, specifically when trained at single epoch tuned 0.0001. results as follows: 99.85%, 100%, 0.0075, 0.0047. experimental independently collected gives empirical evidence Vision Transformer (ViT) architecture exhibited compared all other architectures. It achieved flawless score 1.0 standard metrics, Sensitivity, Specificity, Accuracy, F1-score, Precision, Recall, Mathew Correlation Coefficient (MCC), Precision-Recall Area under Curve (AUC PR), Receiver Operating Characteristic Under (ROC-AUC). Therefore, functions reliable diagnostic tool automated categorization individuals chest computed tomography