Page 31 - Shaping ethics, regulation and standardization in AI for health

Shaping ethics, regulation and standardization in AI for health

Tanzania and Great Britain are described, showcasing leadership and strategy in AI applications
used for Dermatology in Tanzania and Great Britain which was discussed during the meetings.
The TG highlights the initiatives, policies, and achievements in both countries concerning the
implementation of AI solutions in the field of dermatology.

Significant progress has been made in exploring existing work on benchmarking AI for
dermatology applications. The group has reviewed relevant literature and research papers,
gaining insights into various benchmarking approaches and methodologies. Future meetings
will focus on developing standardized datasets and protocols, refining evaluation metrics,
and ensuring transparency and reproducibility in benchmarking. Addressing challenges
related to generalizability and scalability of AI models, the group aims to validate AI solutions
across different healthcare settings and geographic regions. Collaborative efforts will shape
a comprehensive benchmarking framework, fostering advancements in AI applications for
dermatology and ultimately leading to improved patient care, early detection, and optimized
treatment strategies.

A�4�4 DEL 10�4: FG-AI4H Topic Description Document for the Topic Group
on falls among the elderly (TG-Falls)

Summary: This TDD specifies a standardized benchmarking for AI-based systems for fall
prevention and management for older people. It reports background, definitions, methods,
and systems related to falls from the most consolidated scientific literature. It offers an overview
of the state of the art of validation and benchmarking of existing systems for fall prediction. It
proposes a methodology for benchmarking AI systems for falls based on systematic reviews of
available datasets and individual participant data meta-analyses (IPD-MA) of the AI systems. It
provides the protocol and preliminary results of such a systematic review and IPD-MA for the
specific subtopic of fall prediction with wearable inertial sensors. See also the commentary
published in Age and Ageing [7].

A�4�5 DEL 10�6: FG-AI4H Topic Description Document for the Topic Group
on malaria (TG-Malaria)

Summary: This TDD specifies a standardized benchmarking for AI-based malaria detection.
It covers all scientific, technical, and administrative aspects relevant for setting up this
benchmarking.

The AI task aims to develop machine learning methods for diagnosing malaria by detecting
plasmodium pathogens in blood smear images. This approach seeks to improve the accuracy
and speed of malaria diagnosis, addressing issues like misdiagnosis and drug resistance due to
the subjective nature of conventional microscopy and the shortage of skilled lab technicians. By
learning good representations of data directly from pixel images, AI-based detection provides a
reliable, fast, and accurate diagnosis, enhancing confidence for lab technicians and potentially
reducing the burden of malaria in sub-Saharan Africa.

Existing AI solutions for malaria diagnosis, such as those developed at Makerere University's
AI and Data Science lab, have shown improvements in detection accuracies for pathogens
in thick blood smear samples. However, there is no certified AI-based solution for malaria
diagnosis due to limitations like false alarms on undisclosed datasets and the lack of a large,
diverse standardized dataset. Current AI solutions often focus on single detection goals

21