Accurately identify malignant lung nodules intraoperatively using machine learning-based optical biopsy

In a cohort study reported in JAMA network openAzari et al. evaluated whether machine learning-based analysis of intraoperative molecular imaging (IMI) data could accurately and quickly determine the malignancy of indeterminate pulmonary nodules during surgery. This study was conducted to address the persistent challenges in locating small pulmonary nodules and the time, cost, and error-prone nature of intraoperative frozen section analysis. By integrating artificial intelligence (AI)-based image segmentation with statistically derived nomograms, researchers aimed to develop an “optical biopsy” approach that can support real-time surgical decision-making.

Research details

Data were retrospectively analyzed from a prospectively collected database of amorphous pulmonary nodule patients treated at the University of Pennsylvania between 2014 and 2021. A total of 322 patients were included in the study, 279 of whom had complete clinical and imaging data suitable for algorithm development and validation. This cohort was predominantly female (62.7%), and all patients were considered to have high-risk nodules that could harbor potentially malignant tumors.

All patients underwent IMI-guided lung surgery to highlight malignant tissue intraoperatively. We quantified the fluorescence intensity from tumor tissue and surrounding normal lung parenchyma and calculated the tumor-to-background ratio (TBR), an important imaging metric. Patients were randomly split into training and validation sets in an 8:2 ratio with extensive cross-validation to reduce selection bias.

The research team developed a machine learning-based image segmentation algorithm that can quickly and reproducibly calculate TBR directly from intraoperative images. This image analyzer was integrated with a nomogram to create a unified optical biopsy algorithm. We then tested this combination system retrospectively in a validation cohort and prospectively in an independent cohort of 61 consecutive patients undergoing IMI-guided lung cancer surgery.

Main findings

In a retrospective analysis, two related nomogram models demonstrated strong discriminatory performance between malignant and benign nodules, with areas under the receiver operating characteristic curves ranging from 0.865 to 0.893. Variables significantly associated with malignancy included smoking history >5 pack-years, elevated ex vivo TBR values, elevated TBR after specimen bisection, and detectable in situ fluorescence. The machine learning-based image analyzer produced TBR measurements comparable to manual calculations, but with significantly less variation and significantly faster processing time.

When integrated into the final optical biopsy algorithm, the system accurately estimated malignancy risk in the validation cohort, correctly classifying all benign lesions and 96% of invasive adenocarcinomas. Known technical limitations, such as fluorescence attenuation by blood products, resulted in misclassification in a small number of cases.

In a prospective cohort, the algorithm showed a sensitivity of 93.8%, specificity of 100%, positive predictive value of 100%, and negative predictive value of 71%. Of note, the algorithm produced results in less than 2 minutes on average, compared to the average 34 minutes required for frozen section analysis. False-negative results mainly occurred in patients with heavy smoking history and significant anthrax, resulting in increased background fluorescence and decreased contrast between tumor and normal tissue.

The authors concluded: “In this cohort study of patients with irregularly shaped lung nodules, intraoperative imaging data analyzed by AI accurately determined whether the nodules were malignant. This may improve the diagnostic challenges encountered during surgery.”

Dr. Feredun Azariof the Cleveland Clinic Foundation Heart, Vascular and Thoracic Institute in Cleveland, Ohio, is the corresponding author of this paper. JAMA network open article.

Disclosure: For full research author disclosures, please visit jamanetwork.com.