Generative Adversarial Networks for Stain Normalisation in Histopathology
統計
デジタル病理学画像の可視的変動がAIモデルへの影響を示す。
Reinhard, Macenko, Khan, Vahadane等の伝統的手法と比較してGANアプローチが優れていることが多い。
GANアプローチは非生成的手法よりも計算量が高い。
引用
"Stain normalisation aims to standardise the visual profile of digital pathology images without changing the structural content of the images."
"Models have been developed for a wide array of diagnostic and prognostic tasks, with AI researchers aiming to improve the accuracy and efficiency of the interpretation of pathology specimens."
"The most common generative approach is the generative adversarial network (GAN), typically either a single GAN in a supervised setting or multiple GANs in an unsupervised setting."