A novel deep learning segmentation model for organoid-based drug screening.

Area of Science:

• Biotechnology
• Computational Biology
• Oncology

Background:

• Organoids are 3D in vitro cultures from stem cells, valuable for modeling organ development and disease.
• Current organoid growth analysis lacks effective graphic algorithms, hindering drug discovery applications.
• Bladder cancer organoid systems offer a promising platform for therapeutic development.

Purpose of the Study:

• To develop an advanced deep learning model for improved organoid growth analysis.
• To enhance the efficiency and accuracy of organoid-based drug screening.
• To address limitations in current computational methods for organoid imaging.

Main Methods:

• Development of the res-double dynamic conv attention U-Net (RDAU-Net) model.
• Integration of dynamic convolution and attention modules for enhanced feature extraction.
• Training and testing on 200 bladder cancer organoid images across 7 days with drug treatment.

Main Results:

• RDAU-Net demonstrated improved segmentation accuracy (IoU, Dice coefficient) compared to other U-Net variations.
• The model effectively prevented false and missing identifications while maintaining smooth segmentation contours.
• Enhanced feature extraction and multi-scale information utilization improved anti-interference capabilities.

Conclusions:

• The proposed RDAU-Net model significantly improves the efficiency and accuracy of organoid-based drug screening.
• This deep learning approach facilitates high-throughput drug evaluation in cancer research.
• The method offers a robust solution for analyzing complex organoid imaging data.