Geometric Self-Supervised Learning: A Novel AI Framework Towards Quantitative and Explainable Diabetic Retinopathy Detection

Overview

Diabetic retinopathy (DR) is the leading cause of blindness among working -age adults. Early detection is crucial to reducing DR -related vision loss risk but is filled with challenges. Manual detection is labor - intensive and often misses tiny DR -lesions, ne cessitating automated detection. However, existing automated systems are rarely used in clinical practice, solely classifying DR severity into different groups through an uninterpretable black -box process without providing valuable quantitative insight for precision medicine applications. In contrast, a quantitative detection system that identifies individual DR -lesions would overcome these limitations and enable diverse applications in screening, treatment, and research settings, but remains impossible to develop. The reason is that manually annotating diverse lesions is extremely time -consuming and challenging, limiting the amount of reliable data available to train an accurate model. To address this issue, this study presents geometric self -supervised learning, a novel framework for training a deep learning model without any manual annotations as ground truths to detect and segment the four most prevalent types of DR-lesions (i.e., microaneurysms, hemorrhage, hard exudate, and soft exudate) on retinal images, making it possible to utilize the millions of retinal images available for training. Geometric rule -based vision algorithms are utilized to identify and differentiate high -probability normal/abnormal regions and then extract image patches for training a U-net model. This novel framework was extensively verified on two public datasets, significantly outperforming all available studies in detecting and segmenting DR-lesions. It enables self-supervised training of any AI model to detect and segment DR- lesions, and its mechanism is generalizable to other segmentation tasks. TTESSNet: Analysis of Transfer Learning for a TESS Exoplanet Classification Model Jerry Wang Parkland High School, Allentown, PA NASA's Transiting Exoplanet Survey Satellite (TESS) presents an unprecedented volume of space-based photometric observations, with at least ~1,000,000 new light curves generated monthly from full -frame images alone. These data products must be analyzed efficiently and without bias to effectively process and classify light curves and handle the vast data throughput. With these requirements, automated planet candidate classification utilizing deep learning and convolutional neural networks has become an attractive alternative to human vetting. Currently, TESS’s predecessor, the Kepler space telescope, has more research on exoplanet classification models and vastly higher performance. In this project, a transfer learning approach is applied to a TESS exoplanet classification model to bridge the divide and improve the performance and accuracy of TESS exoplanet classification. To conduct the experiment, a light curve preprocessing program was written to process light curves and transit data from the Q1 –Q17 DR 25 TCE table, Kepler Science Data Processing Pipeline, and MIT’s TESS Quick -Look Pipeline. A novel transfer learning approach in model training, where Kepler data is added during training, is analyzed. The results show that the approach noticeably improves the perfo rmance of TESS exoplanet classification when applied to both the testing data set and the ExoFOP catalog. Additionally, potential biases in classified populations are analyzed, along with the possibility of automated TESS vetting and population studies.