The field of mitotic figure classification and detection is moving towards developing more robust and domain-agnostic models. Researchers are exploring various techniques to address the challenges of domain shift, class imbalance, and high morphological variability in histopathological images. The use of modern convolutional architectures, such as ConvNeXt, and training strategies like multi-task learning and domain alignment are showing promising results. Additionally, the incorporation of techniques like style perturbations, attention refinement, and knowledge distillation are helping to improve model performance and reliability. Notable papers in this area include:

• A paper that presents a solution for the MIDOG 2025 Challenge Track 2, achieving robust performance on the diverse MIDOG 2025 dataset through strategic preprocessing and training optimizations.
• A paper that proposes a multi-task neural network for atypical mitosis recognition under domain shift, showing promising performance in a preliminary evaluation conducted on three distinct datasets.