Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session D

Poster A10 - Revolutionizing Cancer Spheroid Viability Analysis with Deep Learning

Friday, October 25, 2024

3:30 PM - 4:30 PM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Rajiv Anne (he/him/his)

Undergraduate Research Assistant
University of Pittsburgh
Pittsburgh, Pennsylvania, United States

Primary Investigator(s)

Yu-Chih Chen, PhD

Assistant Professor
UPMC Hillman Cancer Center/University of Pittsburgh, United States

Introduction: Cancer therapeutics development requires more sophisticated and physiologically relevant models beyond traditional two-dimensional (2D) cell cultures. Three-dimensional (3D) cancer spheroids more accurately simulate the tumor microenvironment and in-vivo conditions. However, despite their advantages, the estimation of cancer spheroid viability faces challenges related to reliability, cytotoxicity, and throughput. Traditional assays rely on invasive and time-consuming methods like fluorescent staining or colorimetric imaging, which can introduce cytotoxicity and affect spheroid viability and behavior, leading to potential inaccuracies. This research aims to overcome these limitations by developing a deep-learning model for label-free, non-destructive real-time evaluation of cancer spheroid viability using phase-contrast images. This novel approach addresses the shortcomings of traditional LIVE/DEAD staining, providing a faster, more reliable, and less toxic alternative to conventional viability assays. By leveraging Convolutional Neural Networks (CNNs), which are particularly effective in image analysis, we aim to revolutionize the assessment of cancer spheroid viability, enhancing the efficiency and accuracy of drug screening processes in cancer research.

Materials and

Methods: We developed a CNN model to assess cancer spheroid viability using phase-contrast imaging. This deep learning algorithm was trained and validated with a comprehensive dataset of phase-contrast images of spheroids treated with eight chemotherapeutic drugs, benchmarked against conventional LIVE/DEAD staining methods. We optimized the network architecture and parameters for maximum prediction accuracy. Phase-contrast imaging, capturing detailed spheroid morphology for viability estimation without fluorescent labels, preserves viability and structure. The model's performance was further tested on novel drugs outside the training dataset to evaluate predictive capability. Inter-cell-line validation confirmed the model's robustness and adaptability across various cancer cell lines. Accuracy was quantified using the R-value, with values close to 1 indicating a strong correlation between predicted and actual viability scores.

Results, Conclusions, and Discussions: The CNN model demonstrated a strong correlation between predicted viability scores and those obtained via LIVE/DEAD staining, with an R-value of 0.9887. This method provided a cost-effective, high-throughput solution for continuous spheroid monitoring, showcasing a robust correlation in the prediction of viability scores across different drugs and treatment conditions. When applied to predict the viability of novel drugs not used in the training set, the model maintained a strong correlation with an R-value of 0.961. Inter-cell-line validation results were equally promising, with the model achieving an R-value of 0.942, indicating its effectiveness across different cell lines. These results underscore the potential of this deep learning approach to revolutionize the assessment of cancer spheroid viability by offering a non-destructive, label-free, and highly accurate alternative to traditional methods.
The findings from this study indicate that the deep learning model offers significant advantages over traditional staining methods, including faster imaging, reduced cost, enhanced image quality, and lower cytotoxicity. This innovative approach promises to improve drug screening efficiency and advance cancer research. Ongoing work tests the model's adaptability to diverse imaging methods, suggesting broad application potential. Future studies will refine the model for greater accuracy and expand its applicability to various cancer types and treatment modalities. Implementing this model in clinical and research settings will revolutionize cancer spheroid viability assessment, providing a powerful tool for oncologists and researchers. This research underscores the potential of integrating deep learning into biomedical imaging and analysis, paving the way for more efficient and accurate therapeutic strategies.

Acknowledgements (Optional):