AI-Powered Cardiac Surgery Support & Dyspnea Case Study

This document introduces CardioVis, an AI-powered copilot designed to revolutionize cardiac surgery by providing real-time intraoperative guidance, objective training analytics, and a federated data platform. It also presents a comprehensive clinical case study focusing on the evaluation and management of acute dyspnea in an adult patient with multiple comorbidities, guiding through diagnosis, lab interpretation, and treatment considerations.

CardioVis: Revolutionizing Cardiac Surgery with AI The field of cardiac surgery faces significant challenges, including high variability in surgical decisions, which contributes to longer aortic cross-clamp times, directly increasing patient mortality by 2% for every additional minute. There's also a substantial patient burden, with conditions like mitral regurgitation affecting 2.5% of the global population. Current training methods are often outdated, relying on observation and simulation without real-time intraoperative support, leading to an estimated $2.2 billion wasted annually in lost operating room time. CardioVis addresses these problems with a comprehensive AI-powered platform.

The CardioVis solution is built upon three core components: an Intraoperative AI Copilot, which delivers real-time, step-by-step visual and voice guidance directly on live surgical feeds through precise annotations, enhancing decision-making and reducing errors. A Training AI Coach analyzes simulation videos to generate objective skill metrics and personalized feedback, aiming to shorten learning curves and elevate surgeon proficiency. Finally, the AUTO-VALVE Data Platform acts as a federated, de-identified backbone for aggregating surgical data, ensuring continuous AI refinement while maintaining robust privacy and scalability.

The technology workflow for CardioVis involves sophisticated data processing: it begins with frame extraction and synchronization from various sources like CT scans, transesophageal echocardiography (TEE), and surgical video streams to create unified multimodal input. Advanced AI models, such as SAM and nnUNet, are then used for AI labeling, automatically identifying anatomy, surgical tools, and procedural phases with high accuracy to provide contextual awareness. Real-time delivery is achieved through heads-up display (HUD) overlays and concise, noise-robust voice prompts designed to seamlessly support surgeons without disrupting their workflow. The system also features continuous model improvement through active learning loops that incorporate new data to enhance performance and adaptability across diverse surgical scenarios.

CardioVis operates on a SaaS (Software as a Service) plus platform licensing business model. Revenue streams include annual AI Copilot licenses and integration fees for hospitals, subscriptions to the AI Coach Mode for training programs, and API access for industry and device OEMs for surgical video analytics. The company has demonstrated significant traction, including an IRB-approved pilot study at University Medical Center, Ho Chi Minh City, involving 24 de-identified mitral valve cases. Benchmarked results show a Dice score of ≥ 0.83 for valve segmentation and real-time inference with only 250 ms latency. CardioVis is supported by Beth Israel Deaconess Medical Center and PASSIO AI Lab, and has secured a $100k contract with Corcym.

Dyspnea Case Discussion: Acute Presentation and Evaluation This educational session aims to teach medical students how to develop a multi-systemic differential diagnosis for adult patients presenting with acute dyspnea, identify common diagnostic studies for evaluation, and correlate pathophysiology with patient presentation and potential treatment options for common causes. The presented case involves a 65-year-old man with a significant medical history including hypertension, hyperlipidemia, coronary artery disease, type 2 diabetes, and chronic kidney disease stage 3a. He presents to the emergency room with rapid onset shortness of breath that began the previous night, marked by an inability to walk more than 20 feet without dyspnea, orthopnea (sleeping in a recliner), paroxysmal nocturnal dyspnea (waking up short of breath), worsening leg swelling, and a frequent cough. He denies chest pain, palpitations, fevers, or chills, and reports no prior similar symptoms. His medications include Aspirin, Lisinopril, Hydrochlorothiazide, Amlodipine, Atorvastatin, and Metformin. Social history is notable for a 30-pack-year former smoking history.

Upon physical examination, the patient's vital signs were concerning: atrial fibrillation with a heart rate of 95, elevated blood pressure of 195/110, tachypnea at 25 breaths/min, and hypoxemia with oxygen saturation of 85% on room air. He was obese (BMI 38) and spoke in shortened sentences. His neck examination revealed an elevated Jugular Venous Pressure (JVP) of 15cm. Cardiovascular exam found a regular rate and rhythm, normal S1 and S2, an S4 heart sound, but no murmurs or rubs. Respiratory examination revealed dullness at the bases with crackles extending up to the mid-lung fields bilaterally, without wheezes or rhonchi. Extremities were warm and well perfused, with bilateral 2+ edema to the shins.

Laboratory data provided crucial insights: his electrolytes (Na 141, K 4.3, Cl 100, CO2 25) were within normal limits. However, kidney function was impaired with BUN 25 and Creatinine 1.7, consistent with his known CKD. Cardiac biomarkers showed an elevated and trending Troponin (150 → 180 → 175), indicating acute myocardial injury. His NT-proBNP was markedly elevated at 1350, strongly supporting a diagnosis of heart failure. A normal D-Dimer of 200 effectively ruled out pulmonary embolism as the primary acute cause of dyspnea in this context. Complete blood count results were within expected ranges (WBC 5, RBC 9, Platelets 300).

Given the comprehensive clinical picture, including the patient's comorbidities, acute onset of symptoms, classic signs of fluid overload (orthopnea, PND, elevated JVP, crackles, edema), and elevated cardiac biomarkers (Troponin and NT-proBNP), the most likely diagnosis is acute decompensated heart failure, possibly precipitated by an acute coronary event or worsening renal function. Acute treatment goals would focus on rapidly improving oxygenation, reducing cardiac preload and afterload (e.g., with oxygen therapy, diuretics, vasodilators), and managing any acute coronary syndrome or arrhythmia. Long-term prevention involves optimizing heart failure medical therapy (e.g., ACE inhibitors/ARBs, beta-blockers, MRAs, SGLT2 inhibitors), aggressive management of his hypertension, diabetes, and hyperlipidemia, smoking cessation, dietary modifications (e.g., low sodium), and close follow-up with cardiology and nephrology to prevent future exacerbations.

Browse public study notes