Hyperspectral imaging techniques (HSI) do not require contact with patients and are non-ionizing as well as non-invasive. As a consequence, they have been extensively applied in the medical field. HSI is being combined with machine learning (ML) processes to obtain models to assist in diagnosis. In particular, the combination of these techniques has proven to be a reliable aid in the differentiation of healthy and tumor tissue during brain tumor surgery. ML algorithms such as support vector machine (SVM), random forest (RF) and convolutional neural networks (CNN) are used to make predictions and provide in-vivo visualizations that may assist neurosurgeons in being more precise, hence reducing damages to healthy tissue. In this work, thirteen in-vivo hyperspectral images from twelve different patients with high-grade gliomas (grade III and IV) have been selected to train SVM, RF and CNN classifiers. Five different classes have been defined during the experiments: healthy tissue, tumor, venous blood vessel, arterial blood vessel and dura mater. Overall accuracy ($OACC$) results vary from 60% to 95% depending on the training conditions. Finally, as far as the contribution of each band to the $OACC$ is concerned, the results obtained in this work are 3.81 times greater than those reported in the literature.

Hyperspectral imaging (HSI) and machine learning (ML) have been employed in the medical field for classifying highly infiltrative brain tumours. Although existing HSI databases of in vivo human brains are available, they present two main deficiencies. First, the amount of labelled data are scarce, and second, 3D-tissue information is unavailable. To address both issues, we present the SLIMBRAIN database, a multimodal image database of in vivo human brains that provides HS brain tissue data within the 400–1000 nm spectra, as well as RGB, depth and multiview images. Two HS cameras, two depth cameras and different RGB sensors were used to capture images and videos from 193 patients. All the data in the SLIMBRAIN database can be used in a variety of ways, for example, to train ML models with more than 1 million HS pixels available and labelled by neurosurgeons, to reconstruct 3D scenes or to visualize RGB brain images with different pathologies, offering unprecedented flexibility for both the medical and engineering communities.

Cancer remains one of the leading causes of morbidity and mortality in the world, with $\sim 18.1$ million new cases of cancer diagnosed worldwide in 2020. In addition, the International Agency for Research on Cancer estimates that by 2040 the diagnosis of new cases will increase to 27.0 million.¹ In Spain, 1.49% of newly diagnosed cancers in 2022 (4.169 of 280.101) were tumors of the encephalo or nervous system.¹ Identifying pathological tissue from healthy tissue is challenging, especially with aggressive tumors such as grade IV glioblastoma (GB) that have high infiltration capabilities.² In addition, GB has poor long-term survival rates,³ making surgery an unavoidable process to increase patient survival. However, the brain shifting toward the skull opening can result in cerebrospinal fluid leakage and hinder tumor identification due to the alterations in the structure of the surrounding tissue, rendering preoperative imaging inadequate for intra-operative conditions.⁴ Therefore, intraoperative tools for brain tumor surgery are essential to help neurosurgeons in delineating and locating the tumor. Neuronavigators are precise instruments that facilitate the real-time monitoring of surgical interventions through the use of magnetic resonance imaging (MRI) or computed tomography scans conducted prior to the procedure. Nonetheless, they present limitations in pinpointing its exact location once the brain is exposed.⁵ Solutions that have been developed to address the issues associated with neuronavigators include the use of intraoperative MRI. This has the advantage of being able to locate the tumor after the craniotomy, thereby solving the problem of brain shift. Nevertheless, it should be noted that the use of MRI increases the time required for surgery and that it requires the use of specific equipment during the surgical procedure.^6,7 Another tool that is not affected by the brain shift issue that operates in real time at a low cost is the intraoperative ultrasound. However, the data must be interpreted by experienced users since the images are of low resolution.^8,9 Fluorescent tumor markers from add-on agents such as 5-aminolevunilic acid are able to deliver highly accurate intraoperative tumor margin detection with a rapid refresh rate. Yet, these are invasive methods since they require the injection of the agent¹⁰ into the patient and present a limited ability to define the tumor margin during surgery for low-grade gliomas.¹¹

Hyperspectral imaging (HSI) has a crucial role in material classification tasks. The primary type of sensors in this field are linescan sensors, offering a high spatial and spectral resolution and improving the quality of the captured hypercubes. However, these cameras have inherent problems such as the need of dedicated hardware to move the camera over the scene and the necessity of pre-processing the raw captures in order to obtain the actual hypercube. This work makes an analysis and an adaptation of stitching techniques to apply them to the medical area. Also, it includes different methods to post-process the hypercubes obtained and it analyzes the SURF algorithm to improve processing times. The results obtained show that a stitching technique based on SURF is suitable for the medical area.

Uyperspectral imaging (HSI) has been adopted during the last years in different applications where material classification plays an important role. This favoured the improvement and development of new HS sensors, leading to HS snapshot cameras, where the sensor is able to acquire images in a realtime video fashion, reducing the spatial and spectral resolution for these cameras. However, although these cameras are able to acquire RAW HS images at high frames per second (FPS) rates, it is necessary to pre-process them to obtain an actual HS image and then obtain a classification map. This work addresses the development of an embedded CPU+GPU HS video classification system able to acquire and classify HS images in real-time video, i.e., at more than 25 FPS. Also, it includes a comparison where three different CPU+GPU embedded platforms are tested: NVIDIA Jetson Nano, NVIDIA Jetson TX1 and NVIDIA Jetson TX2. The results obtained show the feasibility of a HS video classification system using these embedded platforms.