Scholarlink

<p>Soft tissue sarcomas (STS) are a morphologically diverse set of tumours arising from the non-visceral soft tissues with variable biologic behavior that is associated with tumour grade. Tumour grade is determined by a combination of 3 parameters measured by a pathologist: Mitotic Count (MC), percentage of necrosis, and degree of differentiation. Of these, MC is the most objective parameter. MC has been targeted by artificial intelligence (AI) models, which have been shown to efficiently obtain a sensitive and highly reproducible mitotic count. To efficiently detect and quantify mitotic figures in canine STS, we developed and validated a deep learning system using convolutional neural networks (OncoPetNet). Validation of the AI model was performed against a data set of 72 individual tumours from 72 dogs (F<sub>1</sub> score: 0.86) and 100 STS from the MIDOG++ data set (F<sub>1</sub> score: 0.70). The model was evaluated versus manual counts across two studies: study 1 included a single pathologist to mimic the diagnostic situation while controlling for variation among pathologists and study 2 included a group of 9 pathologists to measure inter-pathologist agreement. In both studies, AI-assisted investigation resulted in pathologists observing approximately twice the number of mitotic figures, in approximately half the time of manual counts. Increased MC ultimately resulted in an upward migration of Tumour Grade from Grade 1 to Grade 2 in 15.7% (study 1) to 18% (study 2) of observations and from Grade 2 to 3 in 5.6% (study 1) to 7% (study 2) of observations. With AI-assistance, inter-observer agreement was significantly increased for MC, and MC Scores. AI-assistance also resulted in significant increase in inter-pathologist agreement on Slide Selection. Ultimately, a trend toward increased inter-pathologist agreement in Tumour Grade was also observed; however, this was non-significant indicating variations in other grade parameters like % necrosis and degree of differentiation remain sources of significant inter-pathologist variation. In conclusion, AI-assistance improved efficiency, sensitivity, and inter-pathologist agreement of MC, which is an important step in standardizing STS grading.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Artificial intelligence-assisted mitotic counts improve efficiency, accuracy, and inter-pathologist agreement: a method toward canine soft tissue sarcoma grade standardization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月14日

查看全文
  • Luke Borst,
  • Cindy Bacmeister,
  • Wilson Yau,
  • Alex Aceino,
  • Rose Ranck,
  • Philippe Labelle,
  • Gordon Ehrensing,
  • Fabiano Oliveira,
  • Carl Myers,
  • Jennifer L. Willcox,
  • Zack Ellerby,
  • Richard Haydock,
  • Kylie Taylor,
  • Mark Parkinson,
  • Michael Fitzke,
  • Jodie Gerdin

摘要

Soft tissue sarcomas (STS) are a morphologically diverse set of tumours arising from the non-visceral soft tissues with variable biologic behavior that is associated with tumour grade. Tumour grade is determined by a combination of 3 parameters measured by a pathologist: Mitotic Count (MC), percentage of necrosis, and degree of differentiation. Of these, MC is the most objective parameter. MC has been targeted by artificial intelligence (AI) models, which have been shown to efficiently obtain a sensitive and highly reproducible mitotic count. To efficiently detect and quantify mitotic figures in canine STS, we developed and validated a deep learning system using convolutional neural networks (OncoPetNet). Validation of the AI model was performed against a data set of 72 individual tumours from 72 dogs (F1 score: 0.86) and 100 STS from the MIDOG++ data set (F1 score: 0.70). The model was evaluated versus manual counts across two studies: study 1 included a single pathologist to mimic the diagnostic situation while controlling for variation among pathologists and study 2 included a group of 9 pathologists to measure inter-pathologist agreement. In both studies, AI-assisted investigation resulted in pathologists observing approximately twice the number of mitotic figures, in approximately half the time of manual counts. Increased MC ultimately resulted in an upward migration of Tumour Grade from Grade 1 to Grade 2 in 15.7% (study 1) to 18% (study 2) of observations and from Grade 2 to 3 in 5.6% (study 1) to 7% (study 2) of observations. With AI-assistance, inter-observer agreement was significantly increased for MC, and MC Scores. AI-assistance also resulted in significant increase in inter-pathologist agreement on Slide Selection. Ultimately, a trend toward increased inter-pathologist agreement in Tumour Grade was also observed; however, this was non-significant indicating variations in other grade parameters like % necrosis and degree of differentiation remain sources of significant inter-pathologist variation. In conclusion, AI-assistance improved efficiency, sensitivity, and inter-pathologist agreement of MC, which is an important step in standardizing STS grading.