bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Авг. 21, 2024

Abstract Breast cancer remains one of the most prevalent and deadly cancers worldwide, making accurate diagnosis critical for effective treatment. Histopathological image classification is a key task in medical diagnostics detection. This paper presents state-of-the-art performance histopathological breast using novel approach with Vision Transformer (ViT) model fine-tuned BreakHis dataset. The dataset, comprising 7,909 images across various magnification levels, serves as crucial benchmark evaluating machine learning models this domain. While previous works have explored use ViTs task, our fine-tunes ViT pre-trained on ImageNet Ranger optimizer, achieving unprecedented performance. experimental results show that achieves an accuracy 99.99%, precision 99.98%, recall F1 score specificity 100.00%, false discovery rate (FDR) 0.00%, negative (FNR) 0.02%, positive (FPR) Matthews correlation coefficient (MCC) 99.97%, predictive value (NPV) 99.96%. represents highest ever achieved binary any model, underscoring potential Transformers to substantially enhance diagnostic analysis improve clinical outcomes. Transfer was also performed BACH dataset invasive ductal carcinomas (IDC).

Язык: Английский