Opsyne builds and sells OEM hyperspectral imaging technology that can be integrated into complete industrial automation solutions.
Medical surgical procedures are in certain cases at risk of removing healthy tissue or failing to remove all diseased, cancerous, or unhealthy tissue. This can be for several reasons including poor visibility into the surgical bed due to blood for example, reliance on static medical images that can be challenging to correlate or map onto the surgical bed, or inability to distinguish healthy from diseased tissue. Non-invasive, highly sensitive and specific medical instrumentation can be useful in mapping and distinguishing healthy from unhealthy tissue in real-time as a medical support.
In this study, researchers demonstrate the utility of hyperspectral imaging (HSI) for surgical guidance. The medical team performed an in vivo craniotomy, which is the removal of the skull to access the surgical bed. A Headwall Photonics push-broom hyperspectral sensor operating in the VNIR spectral range from 400 to 1000 nm, was used on a scanning platform covering an effective area of 230 mm.
With the HSI system operated by an engineer, positioned over the exposed brain surface with controlled illumination system, images were captured in two instances, one before the resection of the tumor and one after. In parallel, a biopsy and a histopathological diagnosis were performed for the analysis of the type and grade of the tumor and the surrounding area. Hyperspectral imaging provides the spectral response of each pixel of material across a range of spectral bands. In principle, different materials have unique spectral responses. Healthy tissue will reflect light differently than unhealth tissue. But these are only relative differences. The full analysis requires prediction of the absolute differences. Each pixel of material in the image must be correlated to the properties identified in the biopsy.
The team performing the study prepared a library of classification data to match the pixels of the hyperspectral image, based on their spectral response to their biochemical properties. The dataset was evaluated for accuracy, sensitivity, correlation coefficient and other specific metrics and further computed and classified in two different methods for classification. The results were compared with a machine learning classification method based on the SVM approach. SVM (Support-Vector machine) is a supervised machine learning algorithm for solving classification challenges.
The impact of this study is limited by the number of patients. More classification data is needed to resolve the interpatient variability. More easily obtainable and higher quality hyperspectral images are required during surgical procedures, alongside with higher computational power for the quicker delivery of the results, or the use of snapshot HS cameras which could achieve real-time classification results.
The main contribution of this paper is the achieving of the competitive classification performance at the lowest computational time cost, which is important for the operational use of this technology.
Both processes are significantly less complex than training and labeling steps of the SVM-based classifier. For this particular application, reducing the computational time would allow for personalized results from the spectral data of each patient. This could then be combined with previous patient datasets to develop a classification that takes into account the inter and intra-patient spectral variability.
The advancement of hyperspectral imaging and classification to capture spatial information in real-time could improve the outcomes of the clinical procedures and, hence, improve the patient outcomes and quality of life.
Learn more about how hyperspectral imaging technology can be used to improve insights in specific applications.