Cancer treatment and research communications · 2026
Artificial intelligence-powered evaluation of the pathologic response to preoperative therapy in non-small cell lung cancer.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Ono Akira, Kawata Takuya, Isaka Mitsuhiro, Mori Keita, Shirai Yohei, Otsuka Yujiro, Muramatsu Koji, Tone Kiyoshi, Harada Hideyuki, Ohde Yasuhisa, Sugino Takashi, Takahashi Toshiaki
Аннотация
INTRODUCTION: Quantitative assessment of the pathologic response (pR) in the eyeball approach remains challenging. Few studies have clearly defined viable tumor cell regions (VTRs). We defined VTRs using AE1/AE3 (AE1/3) staining and examined the usefulness of computational pathology (CP) assessment of the pR after preoperative therapy (PT) in non-small cell lung cancer (NSCLC).
METHODS: Among 50 NSCLC patients administered PT at our hospital between October 2002 and April 2021, we selected 31 cases that received PT and evaluated the correlations between AI-based and manual assessments of pR on a slide-by-slide in AE1/3 and hematoxylin and eosin (HE) stained slides. A total of 406 virtual slides (vs) were generated and divided into a training set (300 vs) and a test set (58 vs) without case duplication. pR was defined as the percent area of VTRs relative to the total tumor bed (TB) area. An experienced pathologist manually assessed the pR on HE slides and annotated TB regions, which were used to train a TB model. A VTR model was first developed on AE1/3 vs using an algorithm (AE1/3-trained VTR model), and the resulting labels were transferred to the HEvs to train a HE based VTR model (HE-trained VTR model).
RESULTS: In the test set, AI assessed pR was strongly correlated with the manually assessed pR in the AE1/3 (Spearman's ρ, 0.87, 95%CI, 0.74-0.94) and HE vs. (ρ 0.71, 95%CI, 0.49-0.86).
CONCLUSION: The AI-based CP platform appears to be feasible for quantitative evaluation of the pR following PT in cases of NSCLC.
METHODS: Among 50 NSCLC patients administered PT at our hospital between October 2002 and April 2021, we selected 31 cases that received PT and evaluated the correlations between AI-based and manual assessments of pR on a slide-by-slide in AE1/3 and hematoxylin and eosin (HE) stained slides. A total of 406 virtual slides (vs) were generated and divided into a training set (300 vs) and a test set (58 vs) without case duplication. pR was defined as the percent area of VTRs relative to the total tumor bed (TB) area. An experienced pathologist manually assessed the pR on HE slides and annotated TB regions, which were used to train a TB model. A VTR model was first developed on AE1/3 vs using an algorithm (AE1/3-trained VTR model), and the resulting labels were transferred to the HEvs to train a HE based VTR model (HE-trained VTR model).
RESULTS: In the test set, AI assessed pR was strongly correlated with the manually assessed pR in the AE1/3 (Spearman's ρ, 0.87, 95%CI, 0.74-0.94) and HE vs. (ρ 0.71, 95%CI, 0.49-0.86).
CONCLUSION: The AI-based CP platform appears to be feasible for quantitative evaluation of the pR following PT in cases of NSCLC.