Federated learning & AI aid hospital cancer detection

Federated Learning and AI Help Medical Facilities Detect Cancer Better. A panel of specialists from top US research institutes and hospitals is evaluating the efficacy of federated learning and AI-assisted annotation in developing AI models for tumor segmentation using NVIDIA-powered Federated learning.

Federated Learning: What Is It?

Federated learning is a technique for building more accurate, widely applicable AI models that are trained on data from multiple data sources without jeopardizing data security or privacy. It makes it possible for multiple businesses to collaborate on the development of an AI model without ever letting sensitive data out of their systems.

"Using Federated learning to create and test models at multiple locations simultaneously is the only practical way to stay ahead." It's a very practical tool.

For their most recent project, the team which included partners from a variety of academic institutions, including Case Western, Georgetown, Mayo Clinic, University of California, San Diego, University of Florida, and Vanderbilt University used NVIDIA FLARE (NVFlare), an open-source framework with robust security features, advanced privacy protection techniques, and a flexible system architecture.

Through the NVIDIA Academic Grant Program, the committee received four NVIDIA RTX A5000 GPUs, which were then distributed among the participating research institutes so that the committee members may set up their workstations for Federated learning. Further partnerships used NVIDIA GPUs in cloud and on-premises server environments to show off NVFLare's flexibility.

AI with Federated Learning

Remote Collaboration Amongst Parties

Federated learning makes it possible to construct and validate more accurate and widely applicable AI models from a range of data sources while lowering the risk of endangering data security or privacy. It allows a collection of data sources to be used to develop AI models without the data ever leaving the designated area.

Features

Privacy-Preserving Algorithms

Every change made to the global model is kept private with the use of privacy-preserving strategies from NVIDIA FLARE, and the server is unable to discover any training data or reverse-engineer the weights that users enter.

Procedures for Instruction and Assessment

Among the integrated workflow paradigms that use local and decentralized data to preserve the relevance of models at the edge are learning algorithms for FedAvg, FedOpt, and FedProx.

Extensive Management Tools

Management tools offer safe provisioning using SSL certificates, orchestration through an admin site, and TensorBoard visualization for Federated learning experiments.

Allows for Popular ML/DL Frameworks

The SDK can help you include federated learning into your current workflow. It works with PyTorch, Tensorflow, and even Numpy and has a customizable design.

Extensive API

Its extensive and open-source API allows researchers to develop new federated workflow strategies, creative learning, and privacy-preserving algorithms.

Reusable Building Components

Federated learning experiments can be easily carried out with the help of NVIDIA FLARE, which provides an example walkthrough and reusable building piece.

Cracking the Code of Federated Learning

Each of the six partner medical institutes supplied data from approximately fifty medical imaging studies related to renal cell carcinoma, a kind of kidney cancer, for the program. In a Federated learning architecture, model parameters are transmitted from an initial global model to client servers. Each server uses these parameters to set up a localized version of the model that was trained on private corporate information.

Each local model then sends updated parameters to the global model, and they are integrated to form a new global model. The cycle is repeated until the model's predictions are no longer improving with every training round. The team experimented with model topologies and hyperparameters to optimize for training speed, accuracy, and the number of imaging tests required to train the model to the necessary level of precision.

AI-Assisted Annotation with NVIDIA MONAI

The first part of the study involved manually labeling the model's training set. The team's next move is to evaluate the model's performance with training data segmented using AI against traditional annotation techniques using NVIDIA MONAI for AI-assisted annotation.

The most challenging conditions for federated learning activities arise when the data is not uniformly distributed. According to Garrett, people just label their data differently, use different imaging equipment, and follow different procedures. "Its goal is to ascertain whether the Federated learning model's overall annotation accuracy is improved by adding MONAI during its second training."

The team is using MONAI Label, an image-labeling tool that enables users to create original AI annotation apps, thereby reducing the time and effort needed to generate new datasets. The segmentations generated by AI will be checked and refined by specialists before being used for model training. The data for both the human and AI-assisted annotation stages are hosted by Flywheel, a leading medical imaging data and AI platform that has integrated NVIDIA MONAI into its offerings.

FLARE from NVIDIA

Federated learning is intended to be used with the flexible, open-source, and domain-neutral NVIDIA Federated Learning Application Runtime Environment (NVIDIA FLARE) SDK. It can be used by platform developers to offer a private, secure solution for distributed multi-party collaboration, and data scientists and academics can adapt the existing ML/DL method to a federated paradigm.

Preserving Confidentiality in Multi-Party Cooperation

By using privacy-preserving algorithms and workflow methodologies, you may create and validate more accurate and widely applicable AI models from a range of data sources while lowering the risk that data security and privacy may be compromised.

Accelerate AI Research

permits researchers and data scientists to alter the existing ML/DL procedure (PyTorch, RAPIDS, Nemo, TensorFlow) to conform to a Federated learning paradigm.