ARK unlocks the potential to self-develop multimodal AI algorithms without the need for coding. It seamlessly integrates the entire process—from data import and labeling to modeling, testing, validation, and publishing—empowering healthcare professionals to streamline workflows, enhance diagnostics, and deliver actionable insights for improved patient outcomes.
AI Model Management
Multi-modal Feature Support
Facilitates the integration of diverse domain features, allowing AI models to analyze various data sources for more comprehensive and accurate predictions.
Segmentation Containers
Provides precision segmentation tools such as the Total Segmentator and MedSam, supporting both automated and manual segmentation tasks for domain images.
Workflow Optimization
Integration with PaxeraHealth Viewer
Enables real-time AI inference and visualization, allowing the overlay of AI-generated insights directly on domain images. This enhances decision-making speed and accuracy for clinicians.
ARK Chatbot
The Chatbot feature offers contextual prompts to guide users according to their current workflow stage, providing support for user guidance, model selection, and scientific searches. It also triggers actionable links based on specific keywords, delivers error messages with alternative solutions, and stores session details—like API calls and configurations—ensuring smooth project creation and model training.
Ark Model Version Control
Model version control with every configuration change, saving details like modification date and time, and displaying these versions in a dropdown on the Model screen in ARK Lab. It allows users to select specific versions for new training, deployment, model comparison, and experiment logging, while ensuring version details are included in model training and evaluations.
Ark Model Experiments Comparison
Provide data scientists with a comprehensive comparison screen that displays performance metrics of multiple trained AI models, including detailed line charts for loss and accuracy over epochs, with distinct colors and styles for easy differentiation, hover labels for experiment and metric names, and support for multiple charts per model output to streamline the selection of the best-performing model.
Support 3D classification model creation
ARK platform now supports 3D data for classification projects, enabling detailed spatial analysis and improved accuracy in complex applications like medical imaging.
ARK Model import
ARK platform now allows seamless import of models in .px format through ARK Manager, making it easy to bring in pre-trained or custom models directly into the platform. This feature streamlines workflows by enabling quick model integration and deployment for various applications.
Public Model library integration (Hugging face)
Integrates Hugging Face API into the ARK platform, enabling users to search, browse, and import models directly, and supports downloading selected models during the data fetch process in ARK Manager to seamlessly incorporate pre-trained capabilities and direct evaluation for data analysis projects.
New Specialized models (zero code model creation)
Integrates advanced AI models to detect a broad spectrum of conditions, including 2D classification models for intracranial hemorrhage, brain tumors, Alzheimer’s disease, lung cancer, kidney tumors, chest pneumonia, thyroid tumors, bone fractures, diabetic retinopathy, and COVID-19. Additionally, In addition to supporting 3D segmentation models for conditions like prostate and pancreatic issues, multiple sclerosis, brain tumors, and renal cell carcinoma (RCC).
Multi-Modal model Creation Support
Multi-Modal capability in ARK enables the creation and management of clinical features, incorporating patient risk factors like history, labs, medications, and vitals. This ensures model compatibility with multi-input features via text and vision encoders, enhancing accuracy and relevance.
GradCam for classification models
Enables the extraction of GradCam layers from the model structure for selection and allows users to visualize model predictions on the evaluation screen, helping to understand which input regions influenced the model’s classification decisions.