A Deep Learning-based Miniature Microscope for Imaging Aging and Alzheimer's Disease Brains - ABSTRACT IdenƟfying Alzheimer’s disease (AD) in its presymptomaƟc stage can allow early intervenƟon and improve paƟent care. Crucially, the AD-induced amyloid/tau pathology is not limited to hippocampal insult or memory loss, but also impairs /disrupts funcƟonal connecƟons that integrate sensory inputs in the cortex. As these sensory deficits often precede the decline of cogniƟve funcƟon in AD paƟents, understanding their characterisƟc altered funcƟonal connecƟvity and neural hyperacƟvity patterns early in the AD cascade has the potenƟal to yield new diagnosƟc biomarkers or therapies. Similarly, blood flow decline and endothelial dysfuncƟon posited by the vascular hypothesis of AD remains underexplored. While the availability of transgenic AD mouse models has created a unique platiorm for invesƟgaƟng how AD pathogenesis can disturb the neurovascular unit (NVU), design limitaƟons of imaging hardware, e.g. bulky PET, MR or SPECT, that are not developed/opƟmized to probe AD onset, make it unfeasible to image AD pathogenesis at the spaƟal scale of the NVU. Specifically, AD insults the NVU on mulƟple fronts including neural, vascular/blood flow change that span from neurons to cortex-wide brain acƟvity changes, which are modulated with uneven sleep cycle fragmentaƟon/disrupted circadian rhythms. In contrast, preclinical imaging methods are restricted to short duraƟons (< 2 h) due to anesthesia use when imaging a small animal AD model with a device >1000× in size (e.g. PET), and even the state-of-the-art AD studies assess AD-inflicted NVU change only once in every 1-2 months, which substanƟally under-samples the Ɵme course of AD onset. Moreover, since no two brains age the same, subject-specificity can also mask AD-related NVU changes when imaged intermittently. Therefore, new imaging technologies that can generate large neuroimaging datasets and covering mulƟple temporal (days-months), spaƟal (neurons-whole cortex), and modal (neural-vascular) scales are needed to characterize the “funcƟonal fingerprints” of cortex-wide NVU disrupƟon during AD onset. Therefore, we are proposing the development of NeuroCube, which is a miniaturized microscope that will enable mulƟmodal, cortex-wide in vivo imaging >30 days during AD onset in mice. Unlike extant microscopes that lack capacity for long-term operaƟon (<3 h), we will use 3D-prinƟng and fabricate NeuroCube as a robust unit for longitudinal imaging amidst the harsh, jolty condiƟons in an animal enclosure (Aim 1A). To avoid photobleaching, we will use low-light levels, obtain images at low-signal-to-noise raƟos (SNR)/resoluƟon (50 µm), and recover high-SNR/resoluƟon (10 µm) images via a deep learning (DL) backed generaƟve adversarial network (GAN) (Aim 1B). To limit oversized data volumes, we will image in 1-min bursts (0.75 GB) per hour, and curate an imaging dataset that characterize cortex-wide changes of AD, together with gender/age-matched controls (Aim 2). We believe that The NeuroCube and publicly shared in vivo AD datasets will become a vital new tool for the broad AD research community. Moreover, NeuroCubes could be widely useful for interrogaƟng cortex-bound dysfuncƟon in aging and other brain disease.
