Segmentation of fish chromosomes in microscopy images: A new dataset

Abstract

The chromosome segmentation is the most important step in automatic karyotype assembling. In this work, we presented a brand new chromosome image dataset and proposed methods for segmenting the chromosomes. Chromosome images are usually low quality, especially fish chromosomes. In order to overcome this issue, we tested three filters to reduce noise and improve image quality. After filtering, we applied adaptive threshold segmentation combined with mathematical morphology and supervised classification methods. Support Vector Machine and k-nearest neighbors were applied to discriminate between chromosomes and image background. The proposed method was applied to segment chromosomes in a new dataset. To enable measure the performance of the methods all chromosomes were manually delineated. The results are evaluated considering the Hausdorff distance and normalized sum of distances between segmented and reference images.

Improving Sickle Cell Disease Classification: A Fusion of Conventional Classifiers, Segmented Images, and Convolutional Neural Networks

“…The accuracy increased from 87.70% with original images to 95.54% with segmented images. This improvement suggests that the segmentation process proposed by [Rodrigues et al 2016] effectively enhances the discriminative features captured by Mo-bileNet, resulting in more accurate classifications. The deep convolutional layers of Mo-bileNet enable it to effectively leverage the spatial information provided by the segmentation, allowing it to capture intricate patterns and structures in the images.…”

“…Subsequently, experiments were conducted to assess the impact of segmenting images using the method proposed by [Rodrigues et al 2016] on feature extraction. The experimental results, shown in Table 2, demonstrate that segmented images improved the performance of both evaluated classifiers.…”

“…The erythrocytes classification was performed using the K-Nearest Neighbors (KNN) algorithm. [Rodrigues et al 2016] presented an approach that focused on image segmentation to distinguish normal, sickle, and other deformed cells from the background in microscopy images. They applied morphological feature extraction, considering a set of nine simple geometric shape features.…”

“…All images are in JPG format and have a resolution of 80 × 80 pixels. Additionally, we utilized the segmented images generated using the method proposed by [Rodrigues et al 2016]. Figure 2 shows some images from each class in the dataset, presenting both the original and segmented versions.…”

“…Accuracy (%) [Rodrigues et al 2016] 94.59 [Gual-Arnau et al 2015] 96. 10 [de Faria et al 2018] 93.67 [da Silva et al 2020] 92.54 [Paz-Soto et al 2020] 98.40 [Alzubaidi et al 2020] 99.98 Our approach 96.80…”