AI Medical Technology

Project Background‍
Dermalyser, developed by AI Medical Technologies, is a revolutionary mobile app enhanced with an external lens that provides high-definition images of various skin changes, including malignant and benign melanomas. The pre-existing functionality of Dermalyser assisted users in identifying potentially malignant melanomas. However, the enhanced project goal was to develop a model capable of classifying melanomas as either in-situ or invasive, providing invaluable insights for surgical planning.

Given the limitations of a tight timeline, both parties were aware of the challenges ahead. They aimed to achieve or surpass the benchmark set by previous studies at Sahlgrenska University Hospital, which reported an Area Under the Curve (AUC) of up to 0.76.

Solution Strategy‍
SoftRobot's approach was multifaceted, integrating advanced data pipeline setup, extensive GPU resources, and meticulous experiment logging to ensure reproducibility and traceability—crucial in the highly regulated medical field. The project involved:
‍
-- Data Pipeline Setup: An intricate system was established to manage and process high-definition images from Dermalyser efficiently. The setup was designed to handle large datasets with multiple GPUs, optimizing the machine learning process for speed and accuracy.
-- Model Development and Validation: The team at SoftRobot developed a computer vision based model, rigorously trained and tested using a comprehensive dataset. The model's performance was initially benchmarked against the historical data from Sahlgrenska, with an aim to meet or exceed an AUC of 0.76.
-- Automated Documentation: Recognising the stringent requirements of regulatory bodies, SoftRobot implemented a high level of automated documentation, ensuring that every step of the model's development and deployment was traceable and reproducible.

Results‍
The model exceeded expectations by achieving an AUC in the mid 0.80's, significantly outperforming all the available benchmarks. This result was also validated by an external test set provided by Sahlgrenska University Hospital, affirming the model's efficacy and robustness. The achievement not only highlighted SoftRobot's technical expertise but also reinforced the viability of AI Medical Technologies' innovative approach to melanoma classification.

Impact
The successful implementation of the model had a profound impact on AI Medical Technologies:

-- Regulatory Approval: The comprehensive documentation and validated results facilitated a quicker than usual approval process by regulatory organizations.
-- Future AI Development: The project laid a solid foundation for future AI-driven initiatives, providing AI Medical Technologies with a scalable and reliable framework for ongoing innovation.
-- Integration and Deployment: The model was seamlessly integrated into the existing Dermalyser app via a well-structured API, ensuring minimal disruption to the end users while enhancing the app's diagnostic capabilities.

Conclusion
The collaboration between AI Medical Technologies and SoftRobot Sweden sets a new benchmark in the application of AI for healthcare diagnostics. The project not only achieved its technical objectives but also demonstrated the potential of AI to significantly enhance medical diagnostics and treatment planning. The professionalism and expertise of SoftRobot Sweden in managing a high-stakes, compliance-focused project were pivotal to the success, paving the way for future advancements in the field of digital health solutions.

This case study serves as a testament to the transformative power of AI in healthcare and the potential for collaborative innovation to overcome complex challenges in medical technology.

‍