HSI for Intersegmental Plane Identification During Sublobar Pulmonary Resections

Lung cancer is the leading cause of cancer-related death worldwide. Due to the generalization of screening strategies, especially for risk populations, an increasing number of lung cancer cases are detected in an early stage. In this regard, lung cancer is also increasingly diagnosed in patients with impaired pulmonary function. For preserving lung function and reducing complication incidence, pulmonary segmentectomies are currently evaluated in this cohort. Thus, the latest version of the German guideline for the prevention, diagnosis, treatment and follow-up of lung cancer recommends segmentectomy for patients with impaired pulmonary function in tumor stage I/II. However, the identification of the intersegmental plane - the key step of segmentectomy - remains challenging. Inaccurate recognition of the intersegmental plane may lead to dysfunction of the remaining lung tissue, mismatching of ventilation or blood flow, or long-term air leakage after surgery, which even requires unplanned secondary surgery. Indocyanine green (ICG) is one the latest evaluated identification methods and is considered as gold standard. Hyperspectral Imaging (HSI) - a newly established intraoperative imaging technique - enables a non-invasive evaluation of tissue perfusion and the discrimination of pulmonary tissue with different tissue perfusion during segmentectomies.

The purpose of this prospective, single-center, non-inferiority IDEAL Stage 2b study is the identification of the intersegmental plane and navigation during sublobar pulmonary resections in lung cancer using Hyperspectral Imaging, the comparison with ICG fluorescence intersegmental plane identification, and the establishment of automatic intersegmental plane navigation using machine learning strategies for intraoperative navigation.

To address this, the intersegmental plane will be detected by both HSI and ICG-fluorescence during pulmonary segmentectomies and the correspondence of the two identification methods will be compared with one another. Using machine learning strategies, the detection of perfused and non-perfused pulmonary tissue and intersegmental plane will be analyzed. Finally, the investigators will study motion tracking for the improvement of future HSI illustrations during surgery.

The hypothesis of this study is that HSI could improve the intraoperative navigation during pulmonary segmentectomies providing as reliable intersegmental plane identification as the gold standard of indocyanine green fluorescence. In this case, an intravenous application of fluorescent dye would not be required anymore for the intersegmental plane identification.

In the case of complex segment resection, a large amount or repeated use of ICG is necessary due to its short pulmonary circulation time. Multiple use of ICG may result in ICG entering the target lung tissue through the bronchial circulation and increases the risk of adverse drug reactions of ICG. In contrast, the advantages of HSI would be a faster and repetitive measurement during surgery. There will be a potential for reducing the total measurement time during intersegmental plane dissection (10 seconds vs. 3 minutes / measurement) and consequently patient's burden. In this context, several studies of HSI-based perfusion measurement during esophageal or colorectal surgery showed already an improved patients' outcome. Furthermore, HSI can be used for surgery on patients with hyperthyroidism or impaired renal or hepatic function.

In order to support this hypothesis, a prospective non-inferiority trial design will be used in this study. To ensure the quality of data acquisition and reporting, the study will be conducted in accordance with the IDEAL reporting guidelines. During pulmonary segmentectomies, the intersegmental plane will be identified by both HSI and ICG fluorescence. The determined HSI intersegmental margin will be benchmarked against the ground truth ICG fluorescence and the feasibility and reproducibility of HSI and ICG mapping will be studied.

Machine learning methods have greatly improved the interpretation of subtle patterns in medical image data. Convolutional neural networks (CNNs) can be considered state-of-the-art for classification and segmentation of medical images. The investigators will extend CNN-based methods for HSI classification and particularly study patch-based differentiation between perfused and non-perfused tissue using ICG and HSI data acquired at the same position. A further challenge is the relatively slow acquisition of HSI (10 seconds/measurement), which makes it prone to motion artifacts, e.g., due to pulsatile motion. To address this, the investigators will study motion tracking, which is also relevant for the future illustration of the segment boundary during surgery.

Machine learning approaches and particularly CNNs allow to directly optimize classifiers based on actual clinical data and the spectral dimension can be handled in a straightforward fashion. Moreover, as a versatile method for image processing, CNNs can also be used for localization and motion compensation during intraoperative imaging, e.g., they can be trained to detect image features and their motion in red/green/blue image streams. This is interesting for the proposed HSI data acquisition, which is based on a sequence of measurements which are sensitive to tissue motion.