Fully Automated End-to-End Neuroimaging Workflow for Mental Health Screening

Nikita Thomas, Akhila Perumalla, Srinivasa Rao, Venkatesan Thangaraj, Keerthi Sravan Ravi, Sairam Geethanath, Hansuk Kim, Girish Srinivasan

Abstract: Growing research has proven that mental illnesses, such as depression and chronic stress, can be better understood through the use of medical imaging. However, these findings fail utilization in real-world clinical settings. The intention of this work is to demonstrate the ability to provide widespread access to medical imaging through computer-assisted technology across various community care providers, thus allowing meaningful use outcomes to population and public health. We have created an end-to-end fully automated approach to mental health screening leveraged by artificial intelligence. This workflow involves four core solutions: a voice-activated user interaction, rapid 3D T1weighted contrast and a resting-state functional acquisition, deep learning-based image analysis, and 3D interactive visualization. Intelligent protocolling is used to obtain optimal pulse sequence parameters to satisfy any time constraint. Rigid registration, k-means tissue classification, and an augmented artificial neural network trained on a public database and Talairach atlas segments out over 60 brain structures with corresponding quantitative measures. Functional MR images are processed via seed-based connectivity to produce functional connectivity maps. Finally, a simplistic user experience implementing text-to-speech, speech-to-text technology coupled with a medical image viewer gives interactive passage to medical imaging results. While various tools have emerged to examine both structural and functional MRI within the research space, a simple straightforward, voice-activated interface to perform the scan, automatically process and retrieve meaningful results, that can be accomplished in less than 20 minutes, enables non-specialists to comprehend all relevant information. With this proposed method, insights on a patient's mental health beyond surveys and questionnaires can be determined, such as the quantitative phenotypic impact of stress on the brain, with minimal assistance. This pipeline allows medical imaging to expand beyond radiology for wider population health management protocol adoption.

External IDs:dblp:conf/bibe/ThomasPRTRGKS20