Hyperspectral imaging for real-time intra-surgical brain cancer detection
Gustavo M. Callico
Universidad de Las Palmas de Gran Canaria
Sala de Grados E.T.S.I.Industriales
Brain tumours are among the commonest tumours worldwide with estimated incidences of approximately 3.4 per 100,000. Tumour removal can be a cure for some low grade tumours and can prolong life in more aggressive tumours. However, because brain tumours infiltrate and diffuse into the surrounding normal brain tissue, the surgeon’s naked eye is often unable to distinguish between the tumour and normal brain. Consequently, tumour tissue can unintentionally be left behind during surgery which can later recur. On the other hand, if too much of the tissue is taken out, normal brain is damaged which can lead to permanent disability. Hyperspectral Imaging (HI), collects high resolution spectral information consisting of hundreds of bands across the electromagnetic spectrum, ranging from the ultraviolet to the infrared range. Thanks to this huge amount of information and using a set of complex classification algorithms, it is possible to identify the different materials or substances that compound the image. The HELICoiD (HypErspectraL Imaging Cancer Detection) project is a European FET project that has the goal of developing a demonstrator capable to discriminate with high accuracy between normal and tumour tissues, operating in real-time during neurosurgical operations. This demonstrator can help the neurosurgeons in the process of brain tumour resection, avoiding the excessive extraction of normal tissue and the accidental leaving of small tumour tissues that could cause tumour recurrence. The precise delimitation of the tumour boundaries can improve the results of the brain surgeries. This information will be provided to the surgeon in real-time via different display devices, and in particular by overlaying the conventional images with a simulated colour map which indicates the normal tissues and the tissues affected by tumour cells. The integration of hyperspectral imaging and intraoperative imaged guided surgery systems should have a direct impact on patient outcomes. Potential benefits include: allowing confirmation of complete resection during the surgical procedure, avoiding complications due to "brain shift", and providing confidence that the goals of the surgery have been achieved.
The project has developed a complete system that allows to capture hyperspectral images of the in-vivo brain surface during neurosurgical operations. This HELICoiD demonstrator is capable to acquire two hyperspectral datacubes, one in the VNIR range (covering from 400 nm to 1000 nm) and another one in the NIR range (covering from 900 nm to 1700 nm). Using this demonstrator, a database of 33 hyperspectral images of both spectral ranges from 22 different patients has been generated. From this hypercubes, two different datasets (VNIR and NIR) have been created with pixels labelled as tumour and normal tissue. A classification framework based on a supervised classifier employed to distinguish between the different classes of the labelled samples has been developed. Preliminary results of the classification algorithms offer high accuracy (over 95%) in the discrimination between normal and tumour tissues using labelled samples. Additionally, classification maps of the whole brain surface hyperspectral image have been generated to indicate the precise localization of the labelled classes. These classification maps have been assessed by the neurosurgeons that carried out the surgical procedures and it has been considered that this technique to detect the tumour areas in the brain surface is accurate enough and offers promising results.