Foundation models show promise in classifying atypical mitosis in cancer research

Researchers have benchmarked deep learning and vision foundation models for classifying atypical versus normal mitosis, a crucial indicator of tumor malignancy. The study evaluated end-to-end trained models, linear probing, and fine-tuning with LoRA across multiple datasets, including newly introduced ones. Results showed average balanced accuracies up to 0.81 on in-domain data and 0.77 on out-of-domain data, demonstrating the effectiveness of transfer learning techniques for this challenging classification task. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Demonstrates improved accuracy in medical image classification using transfer learning, potentially aiding tumor malignancy assessment.

RANK_REASON Academic paper presenting a benchmark of deep learning models for a specific classification task.

Read on arXiv cs.CV →

• arXiv
• AMi-Br
• AtNorM-Br
• AtNorM-MD
• TCGA
• MIDOG++
• LoRA

• paper
• other