Automatic Segmentation of Microcalcification Clusters

Classification of Microcalcification Clusters in Digital Mammograms Using a Stack Generalization Based Classifier

“…Another point to note is the SVM trained classifier used the trained data partly to estimate the margin, the support vectors, whereas others function classifiers considered the training set to define the decision function, making them more generalizable. When SVM was discarded from the classifier stack the overall classification performance decreased [11], while including SVM resulted in improved classification accuracy (around 90% for the DDSM database) [11], indicating the positive influence of SVM on ensemble learning, where a majority voting scheme was applied for improved generalization and to gain more flexibility to maintain strong prediction performance by averaging out classifiers individual mistakes and thus reducing the risk of over-fitting. For the k-nearest neighbor (kNN) classifier, Figure 10a, the value of k was set to 5 based on cross-validation.…”

“…It is worth mentioning that only topological features were taken in to account to classify MC clusters, rather than concentrating on the morphological and statistical features of the MC clusters. In our previous study [11], we acquire high classification accuracy (100%) for the MIAS database (24 cases) using LOOCV and 10-fold CV with an ensemble classifier. For DDSM, the accuracy was 91% (for LOOCV) and 90.02 ± 1.42% (for 10-fold CV).…”

“…Double reading can improve sensitivity, but a lack of experienced radiologists can be a challenge [ 9 ]. CADx can assist radiologists in detecting abnormalities in an efficient way [ 10 , 11 ] and systems have been developed to provide a second opinion for diagnosis [ 12 ]. Previous studies have developed computerized methods to aid the diagnosis of MC clusters.…”

Recuperación de información por contenido orientada a la clasificación de grupos de microcalcificaciones en mamografías - Protocam

“…La primera consiste en remover el fondo de la ROI; para esto se aplica una transformación morfológica de apertura a la ROI original, usando como elemento estructurante un disco de radio variable con el fin de determinar la información correspondiente al fondo de la imagen (Zhao, 1993). La segunda etapa consiste en la reconstrucción de la ROI a partir de los coeficientes de aproximación obtenidos de una descomposición Wavelet; el objetivo es obtener una ROI donde el contraste de las microcalcificaciones es realzado por el incremento en la intensidad de las regiones donde estas se encuentran, filtrando todo tipo de estructuras y tejidos irrelevantes (Alam et al, 2018;G., 2006). En la tercera etapa se binariza la ROI mediante una técnica de umbralización manual, así se obtiene el número de objetos (microcalcificaciones) que contiene la ROI, además se usa un filtro de tamaño 3 × 3 para eliminar el ruido sobrante.…”

Multiscale connected chain topological modelling for microcalcification classification

“…In the proposed method, 286 mammographic images with microcalcification clusters were used for estimating the robustness of the algorithm, where 136 ROIs were histologically reported 6 as benign and 150 were categorised as malignant. The microcalcifications were segmented using the detection approach developed by Alam et al [49]. The extracted RoIs are initially enhanced using a wavelet-based algorithm.…”