I became interested in applying AI technology using Deep Learning Constitutional Neural Networks to detect Autism from analysis of a facial image. This was based on reading the results of numerous dysmorphology papers on the subject. To that end I have developed the AI program. It is trained on 1000 images of children with Autism gathered via internet searches, and 1000 images of children not having Autism. After training, the program is tested against 220 images not included in the training set. These are evenly divided with 110 being images of children with Autism and 110 being images of children without Autism. The program accuracy in classification of these images is at 93.18%.The attached jpg file shows the training of the network with respect to training and validation loss and accuracy for those familiar with neural network technology. I believe I can achieve an accuracy rate above 98% based on previous AI experience. The limitation on accuracy is due to the limitation of the number and quality of images of children with Autism. Ideally about 3000 high quality  images would be required, and the test data set should also be expanded to 500 images of each type. Further the images I gathered for Autistic children are based on Google's categorization, the accuracy of which I have no way of knowing. At present I have exhausted all the sources I found available for such images.
I am seeking help in the attempt to expand the training and test databases to achieve the fore mentioned level of accuracy. Specifically I am seeking to find an existing database of facial images of children having Autism if such a database exists. If one does not exist I am looking for some means by which parents of Autistic children would submit images in support of this effort.
Any information on recommendations on how I might proceed would be most welcome. If I can achieve the desired level of accuracy my intent would be to publish the results and make the AI program  available to the public via an internet website or alternatively a smart phone application.
I can be contacted directly via e-mail at gpiosenka@cox.net.