Breast cancer is a neoplasm resulting from the multiplication of abnormal cells in the breast tissue that initially form nodules or thickening. The survival of patients with the disease can be highly effective in cases where the diagnosis is early. However, this diagnosis requires the analysis of extensive histopathological breast tissue samples by specialists. This task is considered complex and may require the attention of the pathologist for a long period of time. Fortunately, computational methods, which include digital image processing and deep learning, can help quickly identify tumour malignancies on images from these scans and speed up diagnosis. Thus, this work proposes a methodology for the automatic diagnosis of malignant breast lesions on histopathological images based on the mutual measurement of bioinspired texture features (BiT) and deep learning features. Additionally, in this study, we also evaluated the effect of methods of colour normalisation to reduce colour variation in the images. The proposed methodology obtained a result of 92.9% accuracy with the random forest classifier. These numbers show that a combination of BiT with deep features produces better results than the techniques applied individually, and show that the proposed method can be used to compose computer-aided detection systems.

Melanoma remains the leading cause of skin cancer-related deaths worldwide, emphasizing the critical need for early detection to enhance survival rates. Computational methods are pivotal in aiding its diagnosis through medical imaging, necessitating accurate lesion segmentation to facilitate effective interpretation. Our study investigates the comparative efficacy of skin lesion classification with and without segmentation, leveraging pre-trained convolutional neural networks (CNNs) and CapsNet architectures. Findings underscore CNNs’ superiority, highlighting segmentation’s beneficial impact on their classification performance, while CapsNet exhibits a degree of independence from segmentation.

The exponential growth in data volume demands efficient data analysis techniques, with data clustering being crucial but interpretation often posing a challenge. Automated group labeling using decision trees can alleviate this issue. This study compares four decision tree algorithms for automated group labeling, demonstrating that algorithm choice significantly influences performance. CHAID outperforms other algorithms in the Iris and Seeds datasets, while C4.5 excels in the Wine and Glass datasets. The proposed model’s validity is confirmed, highlighting the importance of careful algorithm selection. These findings underscore the potential of automated group labeling models and emphasize the need for further research to refine and expand their applications across various domains.