Background: Skin lesions associated with head and neck carcinomas present a diagnostic challenge. Conventional imaging methods, such as dermoscopy and RGB imaging, often face limitations in providing detailed information about skin lesions and accurately differentiating tumor tissue from healthy skin. Methods: This study developed a novel approach utilizing tissue index images derived from hyperspectral imaging (HSI) in combination with machine learning (ML) classifiers to enhance lesion classification. The primary aim was to identify essential features for categorizing tumor, peritumor, and healthy skin regions using both RGB and hyperspectral data. Detailed skin lesion images of 16 patients, comprising 24 lesions, were acquired using HSI. The first- and second-order statistics radiomic features were extracted from both the tissue index images and RGB images, with the minimum redundancy–maximum relevance (mRMR) algorithm used to select the most relevant ones that played an important role in improving classification accuracy and offering insights into the complexities of skin lesion morphology. We assessed the classification accuracy across three scenarios: using only RGB images (Scenario I), only tissue index images (Scenario II), and their combination (Scenario III). Results: The results indicated an accuracy of 87.73% for RGB images alone, which improved to 91.75% for tissue index images. The area under the curve (AUC) for lesion classifications reached 0.85 with RGB images and over 0.94 with tissue index images. Conclusions: These findings underscore the potential of utilizing HSI-derived tissue index images as a method for the non-invasive characterization of tissues and tumor analysis.
This study investigates the potential of direct prediction of daily extremes of temperature at 2 m from a vertical profile measurement using neural networks (NNs). The analysis is based on 3800 daily profiles measured in the period 2004–2019. Various setups of dense sequential NNs are trained to predict the daily extremes at different lead times ranging from 0 to 500 days into the future. The short- to medium-range forecasts rely mainly on the profile data from the lowest layer—mostly on the temperature in the lowest 1 km. For the long-range forecasts (e.g., 100 days), the NN relies on the data from the whole troposphere. The error increases with forecast lead time, but at the same time, it exhibits periodic behavior for long lead times. The NN forecast beats the persistence forecast but becomes worse than the climatological forecast on day two or three. The forecast slightly improves when the previous-day measurements of temperature extremes are added as a predictor. The best forecast is obtained when the climatological value is added as well, with the biggest improvement in the long-term range where the error is constrained to the climatological forecast error.