Parallel computing and hyper parameter optimization via distributed training for AI models over a multi-GPU environment
Mantiscope IVD is a project containing digital microscopy and cloud-driven systems to enable the storage and analysis of biological samples with the help of scalability through data centers. The system contains two parts. First units are the image acquisition devices which are digital microscopes that digitize the biological samples and upload the images to the cloud for AI development and storage. Second is a cloud driven web page used as an annotation platform comprising images collected through the field and AI models deployed for the analysis. This workflow is used in hematology and pathology that require hard and expertise work for the microscopic specimen evaluation. The system has been tested and developed with the help of Hacettepe University Hematology Department, Gazi University Hematology Department, TOBB ETU pediatric hematology Department, Adana Başkent University pediatric hematology/oncology department with specializing the peripheral blood smear evaluation. The system is ready for a pilot project to scale up the service in 10 different cities (Kilis, Osmaniye, Iğdır, Muş, Nevşehir, Kırşehir, Uşak, Burdur, Bayburt, Erzurum/Oltu) which don’t have hematology experts. The image acquisition devices will be deployed to these cities and 5 hematology experts at the center (Adana, Ağrı, Kayseri, Afyon, Erzurum) will evaluate the results after AI flag generation that shows the possibility of a blood disease. For distributed training of AI algorithms with hyperparameter estimations and multi-GPU based parallelization, the company requires a HPC tool to test our system by aiming to minimize the required time of analysis and finding the best trained model as a Proof of Concept in Türkiye.