Presentation
Human-Machine Function Allocation in an Automated Intraoperative Pathology System
SessionPoster Session 2
DescriptionTopic:
Our work focuses on the critical design considerations for AccessPath, a novel artificial intelligence-driven pathology system that enables access to end-to-end intraoperative tumor margin assessment. The core of our approach involves applying the human factors process of function allocation, as Clegg and colleagues (1989) outlined, to carefully determine which parts of this complex system should be automated and which should be performed by a human operator.
Application:
We are developing AccessPath to address a significant unmet need in cancer care, specifically for localized breast and oral cavity cancers. There were 2.3 billion new breast cancer cases and 685,000 recorded breast cancer deaths across the world in 2020 (Sedata, 2023). There are 300,000 oral cancer cases diagnosed annually, causing about 145,000 deaths per year worldwide (Ferlay et al., 2014; Warnakulasuriya, 2009). Surgical resection is the most effective treatment for these cancers, but confirming the complete removal of all cancerous tissue, known as intraoperative margin assessment, is crucial before surgery concludes (Verkooijen et al., 2005; Kurita et al., 2003). A trained pathologist, essential for this assessment, is often unavailable in many surgical suites across the U.S., particularly in low-resource regions, due to high costs and limited availability (Pradipta et al., 2020). Patients who have positive tumor margins detected following tumor removal surgery have to undergo follow-up procedures to remove the remaining cancer, which includes additional radiation therapy or reoperations, with reoperation rates ranging from 20 to 50% for breast and 20% for oral cavity cancer cases (McCahill et al., 2012; Waljee et al., 2008; Blatt et al., 2022). Funded by the Advanced Research Projects Agency for Health (ARPA-H), AccessPath is being engineered to provide accurate intraoperative margin assessment without the substantial cost burden of a pathologist's physical presence in every surgical suite to reduce reoperation rates. Integrating human factors into developing AccessPath, our work on function allocation is a fundamental step in designing such a system, ensuring tasks are appropriately designated to human or machine operators for maximum efficacy and safety.
Background:
When designing a system intended to substitute the physical presence of a human operator, such as a pathologist, there is often an implicit preference to automate as many functions as possible – an approach known as the "left-over" approach (Hendrick, 1995). However, optimal system design necessitates a more nuanced complementary design approach (Moscoso et al., 1999). This involves carefully considering which specific tasks should be automated and which should remain human-operated, based on many factors including technological limitations, current clinical workflows, clinician team preferences, and other system constraints. Leveraging the comprehensive human factors methodology for function allocation described by Clegg and colleagues (1989), we have meticulously examined the system constraints associated with AccessPath and used this evidence to make informed function allocations. Our research included conducting user studies in current intraoperative margin assessment practices within surgical oncology and pathology to gather essential insights.
Overview of presentation:
Our presentation will detail the systematic application of the function allocation phases outlined by Clegg et al. (1989) to inform our final human-machine allocations. We began by specifying the overarching objectives of AccessPath, detailing minimum requirements and identifying all functions and sub-functions, including any mandatory allocations based on known constraints (Phases I-4). Phase 5, the primary allocation phase, involved semi-structured interviews with clinicians, qualitative analysis of the rich interview data, and organizing this data to designate function allocations. We continue to engage in Phases 6 and 7, which involve consistently evaluating and adjusting these allocations as the AccessPath system develops, ensuring ongoing optimization.
Our user interview study to inform Phase 5 of function allocation, approved by Rice University IRB, involved hour-long semi-structured interviews conducted via Zoom with a diverse group of clinicians through snowball sampling. We interviewed nine oncological surgeons and four pathologists from breast and oral cavity cancer specialties, who shared insights from a cumulative of 21 institutions across Houston (TX), San Antonio (TX), Austin (TX), Rio Grande Valley (TX), Hoisington (KS), Waterloo (IA), Los Angeles (CA), and Rochester (NY). The resource levels of the institutions ranged from rural settings to well-funded cancer research institutions. This extensive input allowed us to capture a broad spectrum of current clinical workflows and preferences. A key finding was the clinicians' views on tissue processing steps: most surgeons expressed that they would not be able to conduct these steps, preferring delegation to a machine or an assistant. Interestingly, pathologists indicated that tissue processing steps could potentially be taught to other clinicians, offering a promising avenue, especially given concerns raised by both groups about fully automating the critical slicing step. Crucially, clinicians expressed general trust in AI to read tissue images and detect positive margins accurately, provided the system's validation method is reasonable and transparent.
Based on these interview results, our key takeaway allocation was nuanced and based on practical clinical workflow considerations. We concluded that human operators should be integral to the tissue processing steps, specifically initial orientation, inking, slicing, and arranging of the specimen, with the machine facilitating only specific sub-functions within these steps. Conversely, the information acquisition and analysis steps, such as generating virtual histopathological results and margin detection, are designated for machine operations. This carefully considered function allocation ensures that AccessPath addresses the critical need for intraoperative margin assessment in underserved areas and leverages the unique strengths of human expertise and advanced AI, fostering a safe, effective, and user-acceptable solution. The importance of our message lies in demonstrating that thoughtful human factors integration is paramount for successful and impactful AI deployment in healthcare, prioritizing patient safety and clinical utility alongside technological innovation.
Research reported here was supported by the Advanced Research Projects Agency for Health (ARPA-H) under Award Number D24AC00296-00. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Advanced Research Projects Agency for Health.
Our work focuses on the critical design considerations for AccessPath, a novel artificial intelligence-driven pathology system that enables access to end-to-end intraoperative tumor margin assessment. The core of our approach involves applying the human factors process of function allocation, as Clegg and colleagues (1989) outlined, to carefully determine which parts of this complex system should be automated and which should be performed by a human operator.
Application:
We are developing AccessPath to address a significant unmet need in cancer care, specifically for localized breast and oral cavity cancers. There were 2.3 billion new breast cancer cases and 685,000 recorded breast cancer deaths across the world in 2020 (Sedata, 2023). There are 300,000 oral cancer cases diagnosed annually, causing about 145,000 deaths per year worldwide (Ferlay et al., 2014; Warnakulasuriya, 2009). Surgical resection is the most effective treatment for these cancers, but confirming the complete removal of all cancerous tissue, known as intraoperative margin assessment, is crucial before surgery concludes (Verkooijen et al., 2005; Kurita et al., 2003). A trained pathologist, essential for this assessment, is often unavailable in many surgical suites across the U.S., particularly in low-resource regions, due to high costs and limited availability (Pradipta et al., 2020). Patients who have positive tumor margins detected following tumor removal surgery have to undergo follow-up procedures to remove the remaining cancer, which includes additional radiation therapy or reoperations, with reoperation rates ranging from 20 to 50% for breast and 20% for oral cavity cancer cases (McCahill et al., 2012; Waljee et al., 2008; Blatt et al., 2022). Funded by the Advanced Research Projects Agency for Health (ARPA-H), AccessPath is being engineered to provide accurate intraoperative margin assessment without the substantial cost burden of a pathologist's physical presence in every surgical suite to reduce reoperation rates. Integrating human factors into developing AccessPath, our work on function allocation is a fundamental step in designing such a system, ensuring tasks are appropriately designated to human or machine operators for maximum efficacy and safety.
Background:
When designing a system intended to substitute the physical presence of a human operator, such as a pathologist, there is often an implicit preference to automate as many functions as possible – an approach known as the "left-over" approach (Hendrick, 1995). However, optimal system design necessitates a more nuanced complementary design approach (Moscoso et al., 1999). This involves carefully considering which specific tasks should be automated and which should remain human-operated, based on many factors including technological limitations, current clinical workflows, clinician team preferences, and other system constraints. Leveraging the comprehensive human factors methodology for function allocation described by Clegg and colleagues (1989), we have meticulously examined the system constraints associated with AccessPath and used this evidence to make informed function allocations. Our research included conducting user studies in current intraoperative margin assessment practices within surgical oncology and pathology to gather essential insights.
Overview of presentation:
Our presentation will detail the systematic application of the function allocation phases outlined by Clegg et al. (1989) to inform our final human-machine allocations. We began by specifying the overarching objectives of AccessPath, detailing minimum requirements and identifying all functions and sub-functions, including any mandatory allocations based on known constraints (Phases I-4). Phase 5, the primary allocation phase, involved semi-structured interviews with clinicians, qualitative analysis of the rich interview data, and organizing this data to designate function allocations. We continue to engage in Phases 6 and 7, which involve consistently evaluating and adjusting these allocations as the AccessPath system develops, ensuring ongoing optimization.
Our user interview study to inform Phase 5 of function allocation, approved by Rice University IRB, involved hour-long semi-structured interviews conducted via Zoom with a diverse group of clinicians through snowball sampling. We interviewed nine oncological surgeons and four pathologists from breast and oral cavity cancer specialties, who shared insights from a cumulative of 21 institutions across Houston (TX), San Antonio (TX), Austin (TX), Rio Grande Valley (TX), Hoisington (KS), Waterloo (IA), Los Angeles (CA), and Rochester (NY). The resource levels of the institutions ranged from rural settings to well-funded cancer research institutions. This extensive input allowed us to capture a broad spectrum of current clinical workflows and preferences. A key finding was the clinicians' views on tissue processing steps: most surgeons expressed that they would not be able to conduct these steps, preferring delegation to a machine or an assistant. Interestingly, pathologists indicated that tissue processing steps could potentially be taught to other clinicians, offering a promising avenue, especially given concerns raised by both groups about fully automating the critical slicing step. Crucially, clinicians expressed general trust in AI to read tissue images and detect positive margins accurately, provided the system's validation method is reasonable and transparent.
Based on these interview results, our key takeaway allocation was nuanced and based on practical clinical workflow considerations. We concluded that human operators should be integral to the tissue processing steps, specifically initial orientation, inking, slicing, and arranging of the specimen, with the machine facilitating only specific sub-functions within these steps. Conversely, the information acquisition and analysis steps, such as generating virtual histopathological results and margin detection, are designated for machine operations. This carefully considered function allocation ensures that AccessPath addresses the critical need for intraoperative margin assessment in underserved areas and leverages the unique strengths of human expertise and advanced AI, fostering a safe, effective, and user-acceptable solution. The importance of our message lies in demonstrating that thoughtful human factors integration is paramount for successful and impactful AI deployment in healthcare, prioritizing patient safety and clinical utility alongside technological innovation.
Research reported here was supported by the Advanced Research Projects Agency for Health (ARPA-H) under Award Number D24AC00296-00. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Advanced Research Projects Agency for Health.
Event Type
Poster Presentation
TimeTuesday, March 244:45pm - 6:15pm EDT
LocationRhinelander Gallery
Medical and Drug Delivery Devices