Joao Batista Florindo

Biomedical image segmentation is a critical task in medical diagnosis and treatment planning, enabling precise delineation of anatomical structures and pathological regions. Despite significant advancements, challenges persist due to the inherent variability, noise and complex morphology present in diverse medical imaging modalities. This paper introduces ConvNeXt‐FD, a novel deep learning architecture for robust biomedical image segmentation, built upon a U‐Net‐like encoder‐decoder framework leveraging the powerful ConvNeXt backbone. Our approach integrates a hybrid loss function combining the Dice coefficient with a boundary‐aware regularization term inspired by a differentiable formulation of fractal dimension, designed to enhance the model's sensitivity to object boundaries and shape fidelity. We rigorously evaluate ConvNeXt‐FD across six distinct biomedical datasets: BUSI (breast ultrasound images), DDTI (thyroid ultrasound images), FluoCells (fluorescent cell images), IDRiD (diabetic retinopathy images for optic disc segmentation), ISIC2018 (skin lesion images) and MoNuSeg (nuclei segmentation). Experimental results demonstrate that ConvNeXt‐FD, particularly when initialized with ImageNet pre‐trained weights, achieves competitive and often superior performance compared to existing state‐of‐the‐art methods across various metrics, including Dice, Jaccard, Accuracy, Sensitivity, Specificity and False Positive Rate. The integration of ConvNeXt as a strong encoder, coupled with the boundary‐aware regularization, proves effective in capturing both high‐level semantic features and fine‐grained boundary details, leading to more accurate and reliable segmentations in challenging biomedical contexts.

This study analyzes the role of individual neurons in neural network performance, focusing on the concept of “rotten” neurons whose removal affects network accuracy. By examining two neural network configurations across the MNIST and SVHN datasets, we demonstrate the diverse impact of neurons, from beneficial to detrimental. Our results reveal that neural network efficiency can be improved by addressing the influence of specific neurons. This research highlights the potential for neuron-level analysis and pruning to enhance neural network optimization.

In recent studies, it has been shown that fluorescence lifetime imaging (FLIM) may reveal intracellular structural details in unstained cytological preparations that are not revealed by standard staining procedures. The aim of our investigation was to examine whether FLIM images could reveal areas suggestive of polymerization in red blood cells (RBCs) of sickle cell disease (SCD) patients. We examined label-free blood films using auto-fluorescence FLIM images of 45 SCD patients and compared the results with those of 27 control persons without hematological disease. All control RBCs revealed homogeneous cytoplasm without any foci. Rounded non-sickled RBCs in SCD showed between zero and three small intensively fluorescent dots with higher lifetime values. In sickled RBCs, we found additionally larger irregularly shaped intensively fluorescent areas with increased FLIM values. These areas were interpreted as equivalent to polymerized hemoglobin. The rounded, non-sickled RBCs of SCD patients with homogeneous cytoplasm were not different from those of the erythrocytes of control patients in light microscopy. Yet, variables from the local binary pattern-transformed matrix of the FLIM values per pixel showed significant differences between non-sickled RBCs and those of control cells. In a linear discriminant analysis, using local binary pattern-transformed texture features (mean and entropy) of the erythrocyte cytoplasm of normal appearing cells, the final model could distinguish between SCD patients and control persons with an accuracy of 84.7% of the patients. When the classification was based on the examination of a single rounded erythrocyte, an accuracy of 68.5% was achieved. Employing the Linear Discriminant Analysis classifier method for machine learning, the accuracy was 68.1%. We believe that our study shows that FLIM is able to disclose the topography of the intracellular polymerization process of hemoglobin in sickle cell disease and that the images are compatible with the theory of the two-step nucleation. Furthermore, we think that the presented technique may be an interesting tool for the investigation of therapeutic inhibition of polymerization.

In recent studies it has been shown that fluorescence lifetime imaging (FLIM) may reveal intracellular structural details in unstained cytological preparations which are not revealed by standard staining procedures. The aim of this investigation was to show whether FLIM images could reveal areas suggestive of polymerization in red blood cells (RBCs) of sickle cell disease (SCD) patients. We examined air-dried unstained blood films using auto-fluorescence FLIM images of 45 SCD patients and compared the results with those of 27 control persons without hematological disease. All control RBCs revealed homogeneous cytoplasm without any foci. Rounded non-sickled RBCs in SCD showed between zero and three small intensively fluorescent dots with higher lifetime values. In sickled RBCs we found additionally larger irregularly shaped intensively fluorescent areas with increased FLIM values. These areas were interpreted as equivalent to polymerized hemoglobin. The rounded non-sickled RBCs of SCD patients with homogeneous cytoplasm were not different from those of erythrocytes of control patients in light microscopy. Yet, variables from the local binary pattern transformed matrix of the FLIM values per pixel, showed significant differences between non-sickled RBCs and those of control cells. In a linear discriminant analysis, using local binary pattern transformed texture features (mean and entropy) of the erythrocyte cytoplasm of normal appearing cells, the final model could distinguish between SCD patients and control persons with an accuracy of 84.7% of the patients.