High-throughput Image-based Clustering of CAR-T/Tumor Cocultures for Rapid and Facile Hit Identification

Chimeric antigen receptor T cell is important because of its potential to treat various diseases. As deep learning continues to advance, using unsupervised methods to classify medical images has become a significant focus because collecting high-quality labeled data for medical images is labor-intensive and time-consuming. Beyond the need for accurate labeling, there is a desire to explore the underlying characteristics of the data, even when labels may be ambiguous or uncertain. To address these challenges, we present a novel approach that combines image clustering with an insightful explanation of how these clusters are formed. Our method employs a U-net combined with a clustering algorithm to segment the dataset into different groups. After clustering, we use various techniques to interpret and elucidate the results. Moreover, our paper introduces a unique dataset focused on cell data, specifically highlighting the developmental patterns of cancer cells and T cells under various experimental conditions. This dataset offers a rich source of information and presents a complex challenge for image classification due to the diversity of conditions and cell behaviors involved. Our study thoroughly compares different architectural models on this new dataset, demonstrating the superior performance of our proposed architecture. Through experimental analysis and ablation studies, we provide substantial evidence of the benefits offered by our architecture, not only in terms of accuracy but also in its ability to reveal deeper insights into the data. This work advances the field of image classification and opens new possibilities for understanding complex biological processes through computer vision.