BME Master's Defense Announcement for Kaushik Subramanian (Y. Kim, advisor) Everyone is invited to attend the public presentation beginning at 3:00 PM. Title: Interplay between Radiomic Features and Image Characteristics in Smartphone Conjunctiva Photographs for Machine Learning Applications Date: 7/16/2025 Time: 3:00 PM Location: MJIS 1083 and Zoom* * https://purdue-edu.zoom.us/j/5593290378?omn=93417438631<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpurdue-edu.zoom.us%2Fj%2F5593290378%3Fomn%3D93417438631&data=05%7C02%7Cbmeroundtable-list%40ecn.purdue.edu%7Ccab8659cc3c14c5bba2b08ddc30152f9%7C4130bd397c53419cb1e58758d6d63f21%7C1%7C0%7C638881132639513058%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=NI%2FDSEYMdK%2F97iKO%2BqKBQXOiyrPkTN2KbmXBfGu6V3w%3D&reserved=0> Thesis Committee: Young Kim, Chair; Yunjie Tong; Steven Steinhubl Abstract: Imaging is a core component of medical diagnostics, and smartphone-based photography has gained traction with recent advances in mobile camera technology and its suitability for point-of-care (POC) use, especially in resource-constrained settings. The conjunctiva-a thin, transparent membrane covering the inner eyelids and the white of the eye-offers a non-invasive window into systemic health. We aim to inform predictive algorithms using such images with spatial and textural patterns obtained through radiomics, a high-throughput computational approach in biomedical imaging . However, the robustness of radiomic features under variable image quality remains uncertain, particularly in uncontrolled POC settings, motivating the need for a systematic evaluation. While prior studies have assessed image quality's impact on radiomics, they have focused on traditional imaging modalities. We investigate the sensitivity of radiomic features to spatial resolution and image noise in smartphone photos of the bulbar and palpebral conjunctiva. Using an image dataset captured across different phones and subjects, we applied controlled distortions to simulate resolution loss (via Bicubic interpolation and Gaussian blur) and noise addition (via synthetic Gaussian and Poisson noise). Features were extracted before and after distortion, and statistical tests identified those that did not significantly change-classified as robust. To support practical applications, we included distortion levels reflecting the lowest resolution and highest noise observed in real-world samples. A subset of features demonstrated robustness to these variations, specifically in resolution and noise. These features provide a promising foundation for guiding feature selection in predictive models leveraging radiomics in smartphone-based clinical photography for POC applications in at-home or resource-limited settings. -- Bmeroundtable-list mailing list Bmeroundtable-list@ecn.purdue.edu https://engineering.purdue.edu/ECN/mailman/listinfo/bmeroundtable-list