Presentation Details
| Optimizing Perioperative Care: AI-Assisted Detection of Pheochromocytomas Prior to Non-Adrenal Surgery Alejandra Riveron1, 2, Elio P.Monsour1, Austin Chao2, Varshith Paduchuri2, Holly O'Brien1, 2, John Morrison1. 1Moffitt Cancer Center, Tampa, FL, USA.2University of South Florida Morsani College of Medicine, Tampa, FL, USA |
Abstract
BACKGROUND: Pheochromocytomas are rare catecholamine-secreting tumors. Induction of anesthesia during non-adrenal surgery may precipitate a hypertensive crisis and hemodynamic instability, particularly in tumors of larger size. We developed an AI-assisted pipeline to identify adrenal nodules with imaging features suspicious for pheochromocytoma before elective surgery, integrating size and attenuation cutoffs to prompt biochemical testing to prevent possible perioperative events.
PURPOSE: We developed an AI-assisted pipeline to identify adrenal nodules with imaging features suspicious for pheochromocytoma before elective surgery, integrating size and attenuation cutoffs to prompt biochemical testing to prevent possible perioperative events. Radiology reports from pre-anesthesia testing (PAT) patients were processed using a Snowflake–Claude large language model (LLM) pipeline. PDFs were parsed via Snowflake’s optical character recognition (OCR) function to extract text, followed by Claude-based inference to identify adrenal nodule presence, size, laterality, and Hounsfield (HU) units. Positive findings were validated through human-in-the-loop auditing by medical providers, with results integrated into a Posit dashboard for continuous monitoring and feedback.
RESULTS: Detected high-risk cases were automatically flagged to the PAT team, allowing timely biochemical screening with plasma metanephrines. This process prevented potential intraoperative hypertensive crises, eliminated last-minute surgical cancellations, and improved interdepartmental workflow efficiency.
CONCLUSIONS: This pipeline enables accurate preoperative recognition of adrenal nodules suspicious for pheochromocytoma, leveraging structured thresholds (> 15 HU, size ≥ 2.5 cm) and automated workflow integration to prompt biochemical testing with plasma metanephrines before anesthesia. By embedding this model into PAT processes, hospitals can successfully mitigate perioperative risk, reduce cancellations, and ensure optimal surgical readiness.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
PURPOSE: We developed an AI-assisted pipeline to identify adrenal nodules with imaging features suspicious for pheochromocytoma before elective surgery, integrating size and attenuation cutoffs to prompt biochemical testing to prevent possible perioperative events. Radiology reports from pre-anesthesia testing (PAT) patients were processed using a Snowflake–Claude large language model (LLM) pipeline. PDFs were parsed via Snowflake’s optical character recognition (OCR) function to extract text, followed by Claude-based inference to identify adrenal nodule presence, size, laterality, and Hounsfield (HU) units. Positive findings were validated through human-in-the-loop auditing by medical providers, with results integrated into a Posit dashboard for continuous monitoring and feedback.
RESULTS: Detected high-risk cases were automatically flagged to the PAT team, allowing timely biochemical screening with plasma metanephrines. This process prevented potential intraoperative hypertensive crises, eliminated last-minute surgical cancellations, and improved interdepartmental workflow efficiency.
CONCLUSIONS: This pipeline enables accurate preoperative recognition of adrenal nodules suspicious for pheochromocytoma, leveraging structured thresholds (> 15 HU, size ≥ 2.5 cm) and automated workflow integration to prompt biochemical testing with plasma metanephrines before anesthesia. By embedding this model into PAT processes, hospitals can successfully mitigate perioperative risk, reduce cancellations, and ensure optimal surgical readiness.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
Content Locked. Log into a registered attendee account to access this presentation.