Organfit: a multi-scale convolutional model with ellipse fitting for organoid identification

Organoids hold significant potential and scientific value in areas such as disease modeling, drug screening, and regenerative medicine. During cultivation, researchers monitor the development of organoids through imaging and apply quantitative methods, such as cell counting, to assess key metrics like drug response. However, existing bright-field image processing techniques often struggle with challenges such as blurred boundaries, low contrast, and noise interference, particularly when organoids overlap. These limitations hinder the accurate segmentation of organoid structures. To address these issues, this paper proposes an innovative approach named OrganFit, which integrates a multi-scale convolutional model for segmenting organoid regions, followed by an optimized elliptical fitting principles to detect and screen junction points in overlapping areas. This method captures the features of organoids with varying sizes and shapes, enabling accurate contour fitting even under conditions of severe overlap and low image contrast. As a result, it effectively distinguishes overlapping organoids and significantly enhances the accuracy and reliability of organoid counting. Experimental results across multiple organoid datasets demonstrate that the proposed method maintains stable performance and robustness, showing adaptable generalization across diverse application scenarios.