Automatic Delineation of Tumor Spheroids in Microscopic Images Using Deep-Learning.

Area of Science:

• Oncology
• Biotechnology
• Medical Imaging

Background:

• Tumor spheroid growth assays are crucial for evaluating cancer therapies in vitro.
• Current analysis relies on threshold-based image segmentation, often requiring extensive manual corrections due to treatment-induced morphological changes.
• This manual process is time-consuming and can introduce variability in results.

Purpose of the Study:

• To develop an artificial intelligence (AI)-based method for automated and accurate delineation of tumor spheroids in growth assays.
• To reduce the reliance on manual delineation and corrections, thereby increasing efficiency and consistency.
• To validate the AI method's performance against manual delineations and assess its impact on treatment effect quantification.

Main Methods:

• A deep learning (DL) framework, nnU-Net v2, was employed to develop the AI model.
• The model was trained and validated on a large dataset of 38,090 microscopic images from mouse pheochromocytoma (MPC) cell spheroids.
• Image data included spheroids subjected to irradiation with particle-emitting radioligands, monitored over 35 days.

Main Results:

• The AI-based method achieved high accuracy, with median Dice Similarity Coefficients (DSC) of 0.979 in the training set and 0.974 in the independent test set.
• A low percentage of delineations (7% and 8%) fell below a DSC of 0.9, indicating robust performance.
• Quantification of treatment effects (SCD50) using AI-based delineations closely matched those derived from manual corrections, demonstrating clinical relevance.

Conclusions:

• The developed AI network provides a fast and accurate solution for tumor spheroid delineation in treatment response assays.
• This automated approach significantly reduces the time required for image analysis, from days to hours per experiment.
• The AI method enhances the efficiency and reliability of in vitro cancer therapy evaluation.