Intraoperative EXamination Using MAChine-learning-based HYperspectral for diagNosis & Autonomous Anatomy Assessment

Summary

Participation Requirements

Description

The intraoperative recognition of target structures, which need to be preserved or selectively removed, is of paramount importance during surgical procedures. This task relies mainly on the anatomical knowledge and experience of the operator. In the setting of minimally invasive surgery, there is a ...

The intraoperative recognition of target structures, which need to be preserved or selectively removed, is of paramount importance during surgical procedures. This task relies mainly on the anatomical knowledge and experience of the operator. In the setting of minimally invasive surgery, there is a reduced tactile feedback and the surgeon's vision is the only clue to discriminate the tissues. Misperception of the anatomy, due to patient-specific pathologic conditions and/or to the surgeon's inexperience, can lead to an increased risk of iatrogenic injury of critical anatomical structures and can have devastating consequences. Hyperspectral imaging (HSI) represents a promising technology that combines a photo camera to a spectrometer and that is able to perform a real-time optical scanning over a large area, in a contrast-free manner, providing both spatial and spectral information, generated by the tissue/light interaction. The technology is based on the use of reflectance spectroscopic imaging measurements. The measurement consists in the irradiation of white light on the area (normal halogen lamps, in non-harmful intensity) and the recording of the remitted spectral intensities from the area in the form of remission spectra. The optical interaction (scattering, absorption) of the incident light with the various components (including the depth) of the target material (e.g. biological tissues) alters the spectral distribution of light so that the remitted light carries information about the current material or tissue composition and physiology (e.g. perfusion). HSI is an already established method of objectively classifying image information in a number of scientific fields (e.g. remote sensing), which was first applied in the area of human medicine about 15 years ago. Because of the intrinsic advantages of non-destructive sample collection, interfacing possibilities with common optical modalities (microscope, endoscope) and quantitative, examiner independent results, various approaches have been developed in the meantime to harness the potential of hyperspectral imaging in medicine. Its usefulness in the biomedical field has been already extensively prove. It has been previously applied in digestive surgery to quantify intestinal oxygenated hemoglobin during several procedures, or in case of mesenteric ischemia. A number of previous works focused successfully on the ability of HSI to discriminate between normal and tumor tissue, in prostate cancer, colorectal cancer, gastric cancer, glioblastoma, head and neck cancers. In the oncological field, the advances in hyperspectral features classification have been remarkable and lead to the successful use of sophisticated deep learning algorithms. In surgery, the usefulness of HSI camera has been studied to visualize the operative field under difficult bleeding or to detect tumor presence within the resection margins after surgical excision. A japanese group used an HSI system as additional visualization tool to detect intestinal ischemia and also to classify the intraabdominal anatomy. They identified a particular wavelength (756-830 nm) for the differentiation between healthy and less perfused bowel. They also demonstrated that the spleen, colon, small intestine, urinary bladder and peritoneum have different spectral features. This finding might enable in the future HSI-based navigation of the operation field. Our group recently employed HSI as intraoperative tool in the porcine model to quantify perfusion of the organs of the gastrointestinal tract against robust biological markers. Results showed that this technology is able to quantify bowel blood supply with a high degree of precision. Other groups previously attempted to discriminate bile duct from the vessels, esophagus from tracheal tissue, thyroid from parathyroid gland, nerve and ureter from the surrounding tissue. However, those previous works directed on recognizing key anatomical structures were conducted using either simple feature discrimination algorithms or band selection methods. The amount of information obtained after each acquisition, varies according to the camera resolution, but is quite large, therefore machine and deep learning techniques for data classification and feature extraction are required. In a set of controlled experiments in the porcine model, hyperspectral signatures have been successfully used, coupled to machine learning algorithms, to discriminate fine anatomical structures such as nerves or ureters intraoperatively (unpublished data). The i-EX-MACHYNA3 study aims at translating the HSI technology in combination with several deep learning algorithms to differentiate among different classes of human tissues (including key anatomical structures such as BD, nerves and ureters).

Tracking Information