Deep neural network-based detection and segmentation of intracranial aneurysms on 3D rotational DSA

Objective Accurate diagnosis and measurement of intracranial aneurysms are challenging. This study aimed to develop a 3D convolutional neural network (CNN) model to detect and segment intracranial aneurysms (IA) on 3D rotational DSA (3D-RA) images. Methods 3D-RA images were collected and annotated by 5 neuroradiologists. The annotated images were then divided into three datasets: training, validation, and test. A 3D Dense-UNet-like CNN (3D-Dense-UNet) segmentation algorithm was constructed and trained using the training dataset. Diagnostic performance to detect aneurysms and segmentation accuracy was assessed for the final model on the test dataset using the free-response receiver operating characteristic (FROC). Finally, the CNN-inferred maximum diameter was compared against expert measurements by Pearson’s correlation and Bland-Altman limits of agreement (LOA). Results A total of 451 patients with 3D-RA images were split into n = 347/41/63 training/validation/test datasets, respectively. For aneurysm detection, observed FROC analysis showed that the model managed to attain a sensitivity of 0.710 at 0.159 false positives (FP)/case, and 0.986 at 1.49 FP/case. The proposed method had good agreement with reference manual aneurysmal maximum diameter measurements (8.3 ± 4.3 mm vs. 7.8 ± 4.8 mm), with a correlation coefficient r = 0.77, small bias of 0.24 mm, and LOA of -6.2 to 5.71 mm. 37.0% and 77% of diameter measurements were within ±1 mm and ±2.5 mm of expert measurements. Conclusions A 3D-Dense-UNet model can detect and segment aneurysms with relatively high accuracy using 3D-RA images. The automatically measured maximum diameter has potential clinical application value.

A cost-sensitive Bayesian combiner for reducing false positives in mammographic mass detection

Mammography is the most widely used modality for early breast cancer detection. This work proposes a new computer-aided mass detection approach, in which a denoising method called BM3D is first applied to mammograms. Afterwards, using an adaptive segmentation algorithm, images are segmented to suspicious regions of interest (ROIs) and then a classifier is used to understand the features of true positive (TP) and false positive (FP) patterns. In this way, from selected suspicious ROIs, fractal dimension, texture and intensity features are extracted. Subsequently, a discretization approach followed by correlation-based feature selection (CFS) is combined with a genetic algorithm to obtain the most representative features. To neutralize the classifier’s bias in favor of the major class in imbalanced datasets, an oversampling algorithm is used. In the next step, a cost-sensitive ensemble classifier based on a trainable combiner is proposed in order to reduce the number of FP samples. Finally, the presented method is validated on miniMIAS and INBreast datasets. The free-response receiver operating characteristic (FROC) analysis results prove the efficiency of the proposed approach. A sensitivity of 88% and false positive per image (FPpI) of 0.78 for miniMIAS and also a sensitivity of 86% and FPpI of 0.75 for INBreast dataset were obtained.

DIFFERENTIAL ANALYSIS OF BILATERAL MAMMOGRAMS

Computer support for early detection of breast cancer requires a proper mimicking of the way radiologists compare mammographic images; by comparing bilateral (images of the left and right breasts) and temporal images. In this paper, one method for bilateral registration and intensity normalization and two methods for difference analysis are described. The bilateral registration is based on anatomical features and assumptions of how the female breast is deformed under compression. The first method for differential analysis is based on the absolute difference between the registered images while the second method is based on statistical differences between properties of corresponding neighborhoods. The methods are tested on images from the MIAS database (on 100 images with 59 abnormalities distributed over four types) and evaluated by FROC-analysis. The performances of the two methods are similar but the statistical method gives better performance at a lower false positive rate and is better in particular for detecting asymmetrical developments.