Introduction

The development of PENSIEVE-AI was driven by a pressing global issue – the underdiagnosis of cognitive impairment (CI).  Despite the availability of brief cognitive tests, many are not scalable or suitable for individuals with lower literacy, particularly in underserved communities.  Our goal was to create a tool that could overcome these barriers and facilitate early detection of CI across diverse populations.

Research Journey

The journey of PENSIEVE-AI began with a vision I held for creating a scalable, self-administered cognitive test that does not rely on literacy.  I imagined a tool suitable for community settings, particularly in regions like Singapore, where literacy levels vary widely.  This vision took a significant step forward following a serendipitous meeting in 2019, where I met Mr. Christopher Sia, then Assistant Director at the Government Technology Agency of Singapore (GovTech).  We discovered our shared vision of using AI to improve healthcare, which then solidified into a collaborative effort between Singapore General Hospital and GovTech to realize the potential of PENSIEVE-AI.

From the outset, I challenged our team to recruit research participants who were as representative of the community as possible, rather than relying on a convenient sample of hospital patients.  This decision, while ensuring the AI tool's validity for community use, unraveled into a logistical challenge.  We had to address many seemingly minor yet critical issues, such as the approach to recruit community participants who may not be open to research participation, finding community assessment venues, and ensuring support for participants whom we diagnosed with CI.  Special thanks go to Ms. Yanling Tan, our dedicated research coordinator, who tirelessly tackled these challenges during the initial phase, often working through sleepless nights and weekends (Figure 1). 

Figure 1.  A photo taken during one of our 14 community recruitment roadshows

Despite the hurdles, our study was filled with memorable moments.  One such moment was witnessing participants' reactions to the drawing tasks.  Many found the tasks engaging and appreciated the ease of use, reinforcing our belief in the tool's potential for widespread adoption.  A particularly touching moment was when a participant’s wife expressed gratitude for our help in diagnosing her husband with CI and facilitating a referral to Singapore General Hospital for further clinical care.  She has been worried about his memory for years but was unsure how to seek help.  

Implications and Future Directions

PENSIEVE-AI is unique in its use of drawing tasks, which are less dependent on literacy.  It can be self-administered by most participants in under five minutes, making it a scalable solution for detecting CI in the community.  Our deep-learning model, trained on a large, community-representative sample, demonstrated excellent performance in detecting mild cognitive impairment and dementia, comparable to traditional neuropsychological assessments.

While our initial focus was to address a critical need in Singapore, we soon realized that our work could have a broader impact beyond Singapore, potentially benefiting other populations with similarly diverse levels of literacy (e.g. in some Asian and lower- and middle-income countries).  We are excited about future research possibilities, including further validation in different cultural contexts and integration into community health initiatives.

Conclusion

Reflecting on this journey, I am proud of what we have achieved with PENSIEVE-AI.  My hope is that this tool will contribute to better health outcomes by enabling timely diagnosis and intervention for CI in diverse communities.  There are undoubtedly many more untold stories than can fit here, which I can only summarize by saying that, if not by the grace of God, we might not have reached this stage of the work.  

Lastly, I am thankful for my team of co-authors, who through their contributions, embodied the power of interdisciplinary collaboration (Figure 2).  I am also grateful to the Singapore Prime Minister Office’s Smart Nation and Digital Government Office, the senior management of Singapore General Hospital, and all our community partners and participants for their support to make this work possible.

Figure 2.  A group photo taken during our monthly consensus diagnosis meeting

Read the full paper at Nature Communications.