New AI Tool Predicts Complications Before Lung Cancer Surgery

Postoperative complications are common, making accurate and individualized risk stratification essential for perioperative planning and patient counseling. Existing calculators often depend on population averages and subjective judgment, which can miss patient-specific hazards. To help address this challenge, researchers have developed an artificial intelligence tool that estimates complication risk before surgery and provides editable explanations to align predictions with clinical insight.

Developed by the University at Buffalo and Roswell Park Comprehensive Cancer Center, the system is called MIRACLE, for Multimodal Integrated Radiomics and Clinical Language-based Explanation. It is designed to support thoracic surgeons in evaluating complication risk for candidates undergoing lung cancer surgery. The tool provides patient-level estimates intended to improve planning, communication, and shared decision-making.

MIRACLE integrates three preoperative inputs: clinical information such as age, smoking history, lung function, and comorbidities; quantitative features extracted from computed tomography (CT) scans; and plain‑language explanations generated by a large language model (LLM) grounded in surgical oncology literature and clinical guidelines. The model combines prior knowledge with patient-specific data to produce a risk estimate. It also generates a concise summary that surgeons can review and refine to reflect bedside judgment.

A core feature is “intervenability.” Surgeons can add factors not captured in structured data—such as frailty or functional limitations—and the model recalculates risk accordingly. Safeguards are built in to prevent unsupported content or hallucinations, with key variables such as age, gender, and body mass index locked from LLM editing. Clinical inputs are processed only for the session and deleted immediately afterward to support Health Insurance Portability and Accountability Act (HIPAA) compliance and practical deployment.

In a retrospective evaluation of 3,094 lung cancer surgeries performed at Roswell Park between 2009 and 2023, MIRACLE outperformed five machine-learning methods, three open‑source LLMs, and practicing thoracic surgeons. Surgeons correctly identified complications about 45% of the time, while the model achieved 75%–80% sensitivity and approximately 81% accuracy with low false‑positive rates. A panel of thoracic surgeons reported that many explanations aligned with clinical reasoning, while also noting occasional risk overestimation and missed interactions, underscoring the need for clinician oversight.

Findings were presented at the IEEE/CVF Winter Conference on Applications of Computer Vision 2026 in Tucson, and the preprint is available on arXiv. Next steps include prospective, real‑time validation at Roswell Park and potential adaptation of the platform to other surgical specialties.

“This collaboration between UB and Roswell Park is what makes projects like this possible. You need teams that combine deep clinical insight with real technical expertise to build tools that truly serve patients. Otherwise, industry will do it for us, and its priorities aren't always aligned with patient care. When clinicians and scientists work side by side, we can design solutions that put patients' best interests first. That's why this kind of partnership matters,” said Kenneth Patrick Seastedt, MD, a thoracic surgeon at Roswell Park who also serves as an assistant professor of surgery in the Jacobs School of Medicine and Biomedical Sciences at the University at Buffalo.

Related Links
University at Buffalo 
Roswell Park Comprehensive Cancer Center