Products

Modeling Beyond the Eyes

Cardiovascular Screening

We utilize advanced deep learning techniques to revolutionize non-invasive cardiovascular disease prediction through the examination of vascular biomarkers from different ocular regions, including conjunctival and retinal fundus images. Relevant learning approaches adopted with attention-based UNet architectures integrated with a pretrained encoder gave us precision and recall rates above 85% on RFI datasets and above 75% precision with 80% recall on conjunctival image analyses. Furthermore, we are exploring few-shot segmentation with self-supervised transformer architectures to improve adaptability in cases with limited data access, which is a common challenge in this area. The above segmentation lays down a solid basis for morphological assessment and subsequent biomarker extraction, which would be further utilized to connect ocular and cardiovascular biomarkers to evaluate the comprehensive risk of CVD.

Alzheimer’s

Our advanced product pipeline for improving Alzheimer's disease detection combines the power of computer vision and biomedical innovation. Utilizing Optical Coherence Tomography scans, our platform is built on the EfficientNetB6 model, trained on a comprehensive dataset of nearly 84,000 images across categories such as CNV, DRUSEN, DME, and Normal. Achieving an accuracy of 97% on test data, it surpasses traditional models like ResNet50 and sets a new benchmark in precision diagnostics. Designed for scalability and future growth, our platform enables optic nerve segmentation and the tracking of retinal layer changes. This innovation opens new possibilities for early detection and monitoring of neurodegenerative conditions, empowering clinicians with deeper insights to improve patient outcomes.

Mental Health

Utilizing deep learning techniques, we aim to predict major mental health disorders, including Schizophrenia, Conduct Disorder, and Autism. The ABCD study dataset serves as a foundation, integrating advanced methodologies like neuroimaging and genomic analysis for comprehensive model training. The interdisciplinary approach encourages collaboration across diverse fields, with a primary objective of advancing early diagnosis and personalized treatment through deep learning. Additionally, a commitment to transparency is demonstrated by providing an open-source pre-trained model, with anticipated future applications in MRI-based diagnosis to enhance overall diagnostic accuracy.