Alzheimer’s disease is the most prominent form of dementia with 5.7 million patients in the US. Its symptoms include mental decline, forgetfulness, difficulty concentrating, etc. Living with AD has countless cost of living and quality of life implications with the average annual per person expenditure of medicare alone surpassing $24,000. Currently AD diagnosis usually happens in the later stages which is economically ineffective (demonstrated in the ‘2018 Alzheimer’s disease facts and figures report’), is done without biomarker confirmation which can be inaccurate, and can have high costs (MRI/CT scan) which restricts diagnosis costs for some. Additionally there is a massive under-diagnosis of AD in third world countries. Our solution is Lumen, a low-cost smartphone based retinal scan to diagnose AD. To diagnose a patient, the visoScope lens must be clipped onto the smartphone and the subject’s retina must be imaged via the app. This image is then sent to a server and a diagnosis is returned within minutes.The app is still being built and the open source models of the visoScope are still being modified for usage however the algorithm has been completed and shows promising results. Lumen’s hardware is composed of a smartphone and the oDocs visoScope lens. It software is composed of a machine learning algorithm to identify vascular abnormalities indicative of AD. The preprocessing and post processing algorithms for the images were written in openCV, Scikit-image, NumPy, SciPy, and the machine learning component (U-net) was written in TensorFlow and trained on a Tesla K80 GPU.

What inspired you to create your project?

Alzheimer’s disease is a truly global epidemic; but people don’t often realize the importance of an issue until it affects them. When I was in third grade, I remember a vibrant grandpa full of life and joy. He came to visit our immediate family in America every 2 years. Each time he came, he would lose a little bit of his vibrancy and eventually he no longer became the entertaining, charming, funny man that we knew him to be. He was highly reliant on our grandmother and became forgetful and resultantly, irritable. When my family visited him the winter of 2017, he was in complete bed rest with very minimal awareness of what was around him. He passed away later that month. It was hard on everyone in our family and quite frustrating to all of us that there wasn’t anything that we could to about it. A few weeks prior to his passing, I read an article about how AD is quite often diagnosed way too late in the disease continuum and as result patients suffer unnecessary financial and social consequences. I discussed this with Archishman, who told me about how his grandmother had Alzheimer’s disease as well and that he wanted to pursue a project related to AD. We were determined to help in any way we could. Instead of looking into technical or medical problems we started to examine social and financial factors and realized that late diagnosis, lack of biomarker-based confirmation, diagnostic costs, and third world under-diagnosis were issues plaguing the community. We had our objective: devise a simple, low cost method for early stage detection of AD. Since the traditional method of diagnosis was either purely based on neurological/mental status tests and did not utilize medical imaging and had potential for misdiagnosis, or it combine neurological/mental status tests with MRI/CT scans and its high costs limited access to some, our solution needed to bridge the gap between cost and accuracy. We realized that the best way to solve this problem was to take advantage of a cheap and biologically accessible biomarker. After an extensive search, we came across an article by Queens University Belfast that demonstrated that AD can be diagnosed via FUNDUS imaging. This inspired us to create a completely autonomous algorithm to diagnose AD via retinal scan. Over the course of the next few months we put together a team of mentors, in the fields of ophthalmology, computer science, and business and worked relentlessly to complete the algorithm and test it’s effectiveness. As a result, Lumen, turned out to be something we are truly proud of, and hope to share with the world soon.