Segmentation and Analysis of Corpus Callosum in Alzheimer MR Images using Total Variation Based Diffusion Filter and Level Set Method.

2015 Apr 10;51:355-361.

Author information

Abstract

Twenty years ago the author unveiled his inexpensive complex hand model, which reproduced every motion of the human hand. A control system programmed in the Forth language operated its actuators and sensors. Follow-on papers for this popular project were next presented in Texas, Canada and Germany. From this hand grew the authors meter-tall robot (nicknamed ANNIE: Android With Neural Networks, Intellect and Emotions). It received machine vision, facial expressiveness, speech synthesis and speech recognition; a simian version also received a dexterous ape foot. New artificial intelligence features included op-amp neurons for OCR and simulated emotions, hormone emulation, endocannabinoid receptors, fear-trust-love mechanisms, a Grandmother Cell recognizer and artificial consciousness. Simulated illnesses included narcotic addiction, autism, PTSD, fibromyalgia and Alzheimers disease. The author gave 13 robotics-AI presentations at NASA in Houston since 2006. A meter-tall simian robot was proposed with gripping hand-feet for use with space vehicles and to explore distant planets and moons. Also proposed were: intelligent motorized exoskeletons for astronaut force multiplication; a cognitive prosthesis to detect and alleviate decreased crew mental performance; and a gynoid robot medic to tend astronauts in deep space missions. What began as a complex hand model evolved into an innovative robot-AI within two decades.

PMID: 25996739, revised:

“Alzheimers Disease (AD) is a common form of dementia that affects gray and white matter structures of brain. Manifestation of AD leads to cognitive deficits such as memory impairment problems, ability to think and difficulties in performing day to day activities. Although the etiology of this disease is unclear, imaging biomarkers are highly useful in the early diagnosis of AD. Magnetic resonance imaging is an indispensible non-invasive imaging modality that reflects both the geometry and pathology of the brain. Corpus Callosum (CC) is the largest white matter structure as well as the main inter-hemispheric fiber connection that undergoes regional alterations due to AD. Therefore, segmentation and feature extraction are predominantly essential to characterize the CC atrophy. In this work, an attempt has been made to segment CC using edge based level set method. Prior to segmentation, the images are pre-processed using Total Variation (TV) based diffusion filtering to enhance the edge information. Shape based geometric features are extracted from the segmented CC images to analyze the CC atrophy. Results show that the edge based level set method is able to segment CC in both the normal and AD images. TV based diffusion filtering has performed uniform region specific smoothing thereby preserving the texture and small scale details of the image. Consequently, the edge map of CC in both the normal and AD are apparently sharp and distinct with continuous boundaries. This facilitates the final contour to correctly segment CC from the nearby structures. The extracted geometric features such as area, perimeter and minor axis are found to have the percentage difference of 5.97%, 22.22% and 9.52% respectively in the demarcation of AD subjects. As callosal atrophy is significant in the diagnosis of AD, this study seems to be clinically useful.”