Kiana Negarestani’s Post

Why hasn't AI taken off in healthcare? Fundamental to all of it is the lack of supportive infrastructures for its deployment. Let me elaborate. 1. 𝐖𝐞 𝐝𝐨 𝐧𝐨𝐭 𝐡𝐚𝐯𝐞 𝐫𝐨𝐛𝐮𝐬𝐭 𝐝𝐚𝐭𝐚 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬 𝐭𝐨 𝐬𝐮𝐩𝐩𝐨𝐫𝐭 𝐀𝐈 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐚𝐭 𝐬𝐜𝐚𝐥𝐞. #AI algorithms require a steady input of high quality data to perform its needed function. Peruse any AI research paper and you see the amount of effort required to collate, clean and annotate the data just to train and test the algorithm. It simply does not reflect real word conditions. Imagine a high functioning algorithmn that can analyse 100 data points to predict the risk of worsening sepsis in the ICU, but it requires someone to input the data manually for its function. 😖 2. 𝐖𝐞 𝐝𝐨𝐧'𝐭 𝐡𝐚𝐯𝐞 𝐚𝐧 𝐞𝐚𝐬𝐲 𝐰𝐚𝐲 𝐭𝐨 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐞 𝐚𝐧𝐝 𝐝𝐞𝐩𝐥𝐨𝐲 𝐭𝐡𝐞 𝐯𝐚𝐫𝐢𝐨𝐮𝐬 𝐚𝐯𝐚𝐢𝐥𝐚𝐛𝐥𝐞 𝐚𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬. The good news is there are more and more algorithms available commercially out there (mainly in the radiology space). But the current method of actually onboarding a single AI algorithm within a radiology department can take up to 4 - 6 months and a lot of human effort. If it takes a department 6 months to onboard ONE algorithm and then another 6 to evaluate if it works on their local population - can you imagine the willingness of the team to try and onboard any more solutions? 3. 𝐖𝐞 𝐝𝐨 𝐧𝐨𝐭 𝐡𝐚𝐯𝐞 𝐞𝐧𝐨𝐮𝐠𝐡 𝐬𝐤𝐢𝐥𝐥𝐞𝐝 𝐭𝐞𝐚𝐦𝐬 𝐭𝐨 𝐟𝐚𝐜𝐢𝐥𝐢𝐭𝐚𝐭𝐞 𝐭𝐡𝐞 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐨𝐟 𝐀𝐈 𝐚𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬. AI can be complex in not just its initial onboarding but also on its daily maintenance. We need teams who understands these challenges upfront and are able to advise the financial boards of health systems on which algorithms are suitable and those who are not. With dedicated teams comes growing instuitional knowledge and faster processes with each project that the hospital undertakes. 🚫 No team = no learning = no AI. 4. 𝐀𝐈 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐬 𝐮𝐩𝐟𝐫𝐨𝐧𝐭 𝐢𝐧𝐯𝐞𝐬𝐭𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐰𝐞 𝐡𝐚𝐯𝐞 𝐧𝐨𝐭 𝐪𝐮𝐢𝐭𝐞 𝐠𝐨𝐭𝐭𝐞𝐧 𝐫𝐞𝐢𝐦𝐛𝐮𝐫𝐬𝐞𝐦𝐞𝐧𝐭 𝐫𝐢𝐠𝐡𝐭 𝐣𝐮𝐬𝐭 𝐲𝐞𝐭. Teams of skilled people, robust data pipelines and infrastructures to support AI deployment COSTS MONEY. And there is no running from the fact that upfront investment is required in any health systems for these. The massive cost savings that AI may bring has not yet been fully realised given how early we are currently. ____________________________ 𝐘𝐨𝐮 𝐜𝐚𝐧'𝐭 𝐛𝐮𝐢𝐥𝐝 𝐚 𝐜𝐢𝐭𝐲 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐫𝐨𝐚𝐝𝐬. 🛣 We need many of the above if we are to see AI deployed at scale. Fortunately, there is a growing appreciation for the mistakes and challenges that have been made. 👇 What other challenges can you think about? Haris Shuaib | Dr Terence Tan, MBBS, MSc, GDFM, GDOM | James Blackwood CITP MBCS MCMI | Dr Amrita Kumar | Jan Beger | Prof James Teo

Why hasn't AI taken off in healthcare? Fundamental to all of it is the lack of supportive infrastructures for its deployment. Let me elaborate. 1. 𝐖𝐞 𝐝𝐨 𝐧𝐨𝐭 𝐡𝐚𝐯𝐞 𝐫𝐨𝐛𝐮𝐬𝐭 𝐝𝐚𝐭𝐚 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬 𝐭𝐨 𝐬𝐮𝐩𝐩𝐨𝐫𝐭 𝐀𝐈 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐚𝐭 𝐬𝐜𝐚𝐥𝐞. #AI algorithms require a steady input of high quality data to perform its needed function. Peruse any AI research paper and you see the amount of effort required to collate, clean and annotate the data just to train and test the algorithm. It simply does not reflect real word conditions. Imagine a high functioning algorithmn that can analyse 100 data points to predict the risk of worsening sepsis in the ICU, but it requires someone to input the data manually for its function. 😖 2. 𝐖𝐞 𝐝𝐨𝐧'𝐭 𝐡𝐚𝐯𝐞 𝐚𝐧 𝐞𝐚𝐬𝐲 𝐰𝐚𝐲 𝐭𝐨 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐞 𝐚𝐧𝐝 𝐝𝐞𝐩𝐥𝐨𝐲 𝐭𝐡𝐞 𝐯𝐚𝐫𝐢𝐨𝐮𝐬 𝐚𝐯𝐚𝐢𝐥𝐚𝐛𝐥𝐞 𝐚𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬. The good news is there are more and more algorithms available commercially out there (mainly in the radiology space). But the current method of actually onboarding a single AI algorithm within a radiology department can take up to 4 - 6 months and a lot of human effort. If it takes a department 6 months to onboard ONE algorithm and then another 6 to evaluate if it works on their local population - can you imagine the willingness of the team to try and onboard any more solutions? 3. 𝐖𝐞 𝐝𝐨 𝐧𝐨𝐭 𝐡𝐚𝐯𝐞 𝐞𝐧𝐨𝐮𝐠𝐡 𝐬𝐤𝐢𝐥𝐥𝐞𝐝 𝐭𝐞𝐚𝐦𝐬 𝐭𝐨 𝐟𝐚𝐜𝐢𝐥𝐢𝐭𝐚𝐭𝐞 𝐭𝐡𝐞 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐨𝐟 𝐀𝐈 𝐚𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬. AI can be complex in not just its initial onboarding but also on its daily maintenance. We need teams who understands these challenges upfront and are able to advise the financial boards of health systems on which algorithms are suitable and those who are not. With dedicated teams comes growing instuitional knowledge and faster processes with each project that the hospital undertakes. 🚫 No team = no learning = no AI. 4. 𝐀𝐈 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐬 𝐮𝐩𝐟𝐫𝐨𝐧𝐭 𝐢𝐧𝐯𝐞𝐬𝐭𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐰𝐞 𝐡𝐚𝐯𝐞 𝐧𝐨𝐭 𝐪𝐮𝐢𝐭𝐞 𝐠𝐨𝐭𝐭𝐞𝐧 𝐫𝐞𝐢𝐦𝐛𝐮𝐫𝐬𝐞𝐦𝐞𝐧𝐭 𝐫𝐢𝐠𝐡𝐭 𝐣𝐮𝐬𝐭 𝐲𝐞𝐭. Teams of skilled people, robust data pipelines and infrastructures to support AI deployment COSTS MONEY. And there is no running from the fact that upfront investment is required in any health systems for these. The massive cost savings that AI may bring has not yet been fully realised given how early we are currently. ____________________________ 𝐘𝐨𝐮 𝐜𝐚𝐧'𝐭 𝐛𝐮𝐢𝐥𝐝 𝐚 𝐜𝐢𝐭𝐲 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐫𝐨𝐚𝐝𝐬. 🛣 We need many of the above if we are to see AI deployed at scale. Fortunately, there is a growing appreciation for the mistakes and challenges that have been made. 👇 What other challenges can you think about? Haris Shuaib | Dr Terence Tan, MBBS, MSc, GDFM, GDOM | James Blackwood CITP MBCS MCMI | Dr Amrita Kumar | Jan Beger | Prof James Teo

Why hasn't AI taken off in healthcare? Fundamental to all of it is the lack of supportive infrastructures for its deployment. Let me elaborate. 1. 𝐖𝐞 𝐝𝐨 𝐧𝐨𝐭 𝐡𝐚𝐯𝐞 𝐫𝐨𝐛𝐮𝐬𝐭 𝐝𝐚𝐭𝐚 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬 𝐭𝐨 𝐬𝐮𝐩𝐩𝐨𝐫𝐭 𝐀𝐈 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐚𝐭 𝐬𝐜𝐚𝐥𝐞. #AI algorithms require a steady input of high quality data to perform its needed function. Peruse any AI research paper and you see the amount of effort required to collate, clean and annotate the data just to train and test the algorithm. It simply does not reflect real word conditions. Imagine a high functioning algorithmn that can analyse 100 data points to predict the risk of worsening sepsis in the ICU, but it requires someone to input the data manually for its function. 😖 2. 𝐖𝐞 𝐝𝐨𝐧'𝐭 𝐡𝐚𝐯𝐞 𝐚𝐧 𝐞𝐚𝐬𝐲 𝐰𝐚𝐲 𝐭𝐨 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐞 𝐚𝐧𝐝 𝐝𝐞𝐩𝐥𝐨𝐲 𝐭𝐡𝐞 𝐯𝐚𝐫𝐢𝐨𝐮𝐬 𝐚𝐯𝐚𝐢𝐥𝐚𝐛𝐥𝐞 𝐚𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬. The good news is there are more and more algorithms available commercially out there (mainly in the radiology space). But the current method of actually onboarding a single AI algorithm within a radiology department can take up to 4 - 6 months and a lot of human effort. If it takes a department 6 months to onboard ONE algorithm and then another 6 to evaluate if it works on their local population - can you imagine the willingness of the team to try and onboard any more solutions? 3. 𝐖𝐞 𝐝𝐨 𝐧𝐨𝐭 𝐡𝐚𝐯𝐞 𝐞𝐧𝐨𝐮𝐠𝐡 𝐬𝐤𝐢𝐥𝐥𝐞𝐝 𝐭𝐞𝐚𝐦𝐬 𝐭𝐨 𝐟𝐚𝐜𝐢𝐥𝐢𝐭𝐚𝐭𝐞 𝐭𝐡𝐞 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐨𝐟 𝐀𝐈 𝐚𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬. AI can be complex in not just its initial onboarding but also on its daily maintenance. We need teams who understands these challenges upfront and are able to advise the financial boards of health systems on which algorithms are suitable and those who are not. With dedicated teams comes growing instuitional knowledge and faster processes with each project that the hospital undertakes. 🚫 No team = no learning = no AI. 4. 𝐀𝐈 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐬 𝐮𝐩𝐟𝐫𝐨𝐧𝐭 𝐢𝐧𝐯𝐞𝐬𝐭𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐰𝐞 𝐡𝐚𝐯𝐞 𝐧𝐨𝐭 𝐪𝐮𝐢𝐭𝐞 𝐠𝐨𝐭𝐭𝐞𝐧 𝐫𝐞𝐢𝐦𝐛𝐮𝐫𝐬𝐞𝐦𝐞𝐧𝐭 𝐫𝐢𝐠𝐡𝐭 𝐣𝐮𝐬𝐭 𝐲𝐞𝐭. Teams of skilled people, robust data pipelines and infrastructures to support AI deployment COSTS MONEY. And there is no running from the fact that upfront investment is required in any health systems for these. The massive cost savings that AI may bring has not yet been fully realised given how early we are currently. ____________________________ 𝐘𝐨𝐮 𝐜𝐚𝐧'𝐭 𝐛𝐮𝐢𝐥𝐝 𝐚 𝐜𝐢𝐭𝐲 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐫𝐨𝐚𝐝𝐬. 🛣 We need many of the above if we are to see AI deployed at scale. Fortunately, there is a growing appreciation for the mistakes and challenges that have been made. 👇 What other challenges can you think about? Haris Shuaib | Dr Terence Tan, MBBS, MSc, GDFM, GDOM | James Blackwood CITP MBCS MCMI | Dr Amrita Kumar | Jan Beger | Prof James Teo

Why hasn't AI taken off in healthcare? Fundamental to all of it is the lack of supportive infrastructures for its deployment. Let me elaborate. 1. 𝐖𝐞 𝐝𝐨 𝐧𝐨𝐭 𝐡𝐚𝐯𝐞 𝐫𝐨𝐛𝐮𝐬𝐭 𝐝𝐚𝐭𝐚 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬 𝐭𝐨 𝐬𝐮𝐩𝐩𝐨𝐫𝐭 𝐀𝐈 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐚𝐭 𝐬𝐜𝐚𝐥𝐞. #AI algorithms require a steady input of high quality data to perform its needed function. Peruse any AI research paper and you see the amount of effort required to collate, clean and annotate the data just to train and test the algorithm. It simply does not reflect real word conditions. Imagine a high functioning algorithmn that can analyse 100 data points to predict the risk of worsening sepsis in the ICU, but it requires someone to input the data manually for its function. 😖 2. 𝐖𝐞 𝐝𝐨𝐧'𝐭 𝐡𝐚𝐯𝐞 𝐚𝐧 𝐞𝐚𝐬𝐲 𝐰𝐚𝐲 𝐭𝐨 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐞 𝐚𝐧𝐝 𝐝𝐞𝐩𝐥𝐨𝐲 𝐭𝐡𝐞 𝐯𝐚𝐫𝐢𝐨𝐮𝐬 𝐚𝐯𝐚𝐢𝐥𝐚𝐛𝐥𝐞 𝐚𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬. The good news is there are more and more algorithms available commercially out there (mainly in the radiology space). But the current method of actually onboarding a single AI algorithm within a radiology department can take up to 4 - 6 months and a lot of human effort. If it takes a department 6 months to onboard ONE algorithm and then another 6 to evaluate if it works on their local population - can you imagine the willingness of the team to try and onboard any more solutions? 3. 𝐖𝐞 𝐝𝐨 𝐧𝐨𝐭 𝐡𝐚𝐯𝐞 𝐞𝐧𝐨𝐮𝐠𝐡 𝐬𝐤𝐢𝐥𝐥𝐞𝐝 𝐭𝐞𝐚𝐦𝐬 𝐭𝐨 𝐟𝐚𝐜𝐢𝐥𝐢𝐭𝐚𝐭𝐞 𝐭𝐡𝐞 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐨𝐟 𝐀𝐈 𝐚𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬. AI can be complex in not just its initial onboarding but also on its daily maintenance. We need teams who understands these challenges upfront and are able to advise the financial boards of health systems on which algorithms are suitable and those who are not. With dedicated teams comes growing instuitional knowledge and faster processes with each project that the hospital undertakes. 🚫 No team = no learning = no AI. 4. 𝐀𝐈 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐬 𝐮𝐩𝐟𝐫𝐨𝐧𝐭 𝐢𝐧𝐯𝐞𝐬𝐭𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐰𝐞 𝐡𝐚𝐯𝐞 𝐧𝐨𝐭 𝐪𝐮𝐢𝐭𝐞 𝐠𝐨𝐭𝐭𝐞𝐧 𝐫𝐞𝐢𝐦𝐛𝐮𝐫𝐬𝐞𝐦𝐞𝐧𝐭 𝐫𝐢𝐠𝐡𝐭 𝐣𝐮𝐬𝐭 𝐲𝐞𝐭. Teams of skilled people, robust data pipelines and infrastructures to support AI deployment COSTS MONEY. And there is no running from the fact that upfront investment is required in any health systems for these. The massive cost savings that AI may bring has not yet been fully realised given how early we are currently. ____________________________ 𝐘𝐨𝐮 𝐜𝐚𝐧'𝐭 𝐛𝐮𝐢𝐥𝐝 𝐚 𝐜𝐢𝐭𝐲 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐫𝐨𝐚𝐝𝐬. 🛣 We need many of the above if we are to see AI deployed at scale. Fortunately, there is a growing appreciation for the mistakes and challenges that have been made. 👇 What other challenges can you think about? Haris Shuaib | Dr Terence Tan, MBBS, MSc, GDFM, GDOM | James Blackwood CITP MBCS MCMI | Dr Amrita Kumar | Jan Beger | Prof James Teo