1,721,025 research outputs found
An Empirical Comparison of Ideal and Empirical ROC-Based Reject Rules
No full textTwo class classifiers are used in many complex problems in which the classification results could have serious consequences. In such situations the cost for a wrong classification can be so high that can be convenient to avoid a decision and reject the sample. This paper presents a comparison between two different reject rules (the Chow’s and the ROC rule). In particular, the experiments show that the Chow’s rule is inappropriate when the estimates of the a posteriori probabilities are not reliable
A Framework for Multiclass Reject in ECOC Classification Systems
No full textECOC is a diffused and successful technique to implement a multiclass classification system by decomposing the original problem in several two-class problems. In this paper we propose ECOC systems with a reject option carried out through two different schemes. The first one estimates the reliability of the output of the ECOC system and does not require any change in its structure. The second scheme, instead, estimates the reliability of the internal dichotomizers and implies a slight modification in the decoding stage. A final investigation is done on the sequential combination of both methods
Embedding Reject Option in ECOC Through LDPC Codes
No full textError Correcting Output Coding (ECOC) is an established technique to face a classification problem with many possible classes decomposing it into a set of two class subproblems. In this paper, we propose an ECOC system with a reject option that is performed by taking into account the confidence degree of the dichotomizers. Such a scheme makes use of a coding matrix based on Low Density Parity Check (LDPC) codes that can also be usefully employed to implement an iterative recovery strategy for the binary rejects. The experimental results have confirmed the effectiveness of the proposed approach
An effective learning strategy for cascaded object detection
Full text linkTo distinguish objects from non-objects in images under computational constraints, a suitable solution is to employ a cascade detector that consists of a sequence of node classifiers with increasing discriminative power. However, among the millions of image patches generated from an input image, only very few contain the searched object. When trained on these highly unbalanced data sets, the node classifiers tend to have poor performance on the minority class. Thus, we propose a learning strategy aimed at maximizing the node classi-fiers ranking capability rather than their accuracy. We also provide an efficient implementation yielding the same time complexity of the original Viola-Jones cascade training. Experimental results on highly unbalanced real problems show that our approach is both efficient and effective when compared to other node training strategies for skewed classes
Bit Error Recovery in ECOC Systems through LDPC Codes
No full textError Correcting Output Coding (ECOC) is a widely used and successful technique that implements a classification system by splitting a multiclass problem into a set of dichotomies according to a coding matrix. In this paper we propose a new approach for the ECOC systems based on a well-known family of error correcting codes in the Coding Theory: the Low Density Parity Check (LDPC) codes. The goal is twofold: first, to introduce a new coding strategy for determining the code words to be collected in a coding matrix. Second, to exploit the algebraic properties of LDPC codes for recovering the bit errors in the output word and increasing the performance of the classification system. We compare the proposed technique with other commonly used coding strategies on some benchmark data sets achieving very interesting results
A Ranking-based Cascade Approach for Unbalanced Data
No full textIn this paper we present a cascade-based framework
for object detection in which the node classifiers are
trained by a learning algorithm based on ranking in-
stead of classification error. Such an approach is par-
ticularly suited for facing the asymmetry between pos-
itive and negative class, that is a huge problem in ob-
ject detection applications. Other methods focused on
this problem and previously proposed, such as Asym-
Boost, rely on an asymmetric weight updating mech-
anism of the samples based on a parameter k which
estimates the degree of skewing between the classes.
Actually such parameter is difficult to choose and re-
quires a significant tuning activity during the training
phase. On the contrary, our approach is nonparametric
and has demonstrated to provide slightly better perfor-
mance when compared with AsymBoost on a real detec-
tion problem
Convolutional Networks and Transformers for Mammography Classification: An Experimental Study
Full text linkConvolutional Neural Networks (CNN) have received a large share of research in mammography image analysis due to their capability of extracting hierarchical features directly from raw data. Recently, Vision Transformers are emerging as viable alternative to CNNs in medical imaging, in some cases performing on par or better than their convolutional counterparts. In this work, we conduct an extensive experimental study to compare the most recent CNN and Vision Transformer architectures for whole mammograms classification. We selected, trained and tested 33 different models, 19 convolutional- and 14 transformer-based, on the largest publicly available mammography image database OMI-DB. We also performed an analysis of the performance at eight different image resolutions and considering all the individual lesion categories in isolation (masses, calcifications, focal asymmetries, architectural distortions). Our findings confirm the potential of visual transformers, which performed on par with traditional CNNs like ResNet, but at the same time show a superiority of modern convolutional networks like EfficientNet
A Boosting-Based Approach to Refine the Segmentation of Masses in Mammography
No full textIn this paper we present an algorithm for finding an accurate estimate of the contour of masses in mammograms. We assume that a rough estimate of the region containing the mass is known: in particular it is available the location of an area inside the mass (core) and a closed curve beyond which the mass does not extend. The proposed method employs a boosting-based classifier trained on the core and on a background region beyond the external contour, so that it provides an accurate estimate of the mass contour by classifying unlabeled pixels between the core and the external contour. The proposed approach is useful not only for automatic localization of mass contour, but also as a powerful tool during annotation of mammograms, given that an user provides interactively an estimate for the core and the external contour of the mass. The approach has been verified on a set of mammograms showing very encouraging results
Transfer learning in breast mass detection and classification
No full textCovid-19 infection influenced the screening test rate of breast cancer worldwide due to the quarantine measures, routine procedures reduction, and delay of early diagnosis, causing high mortality risk and severity of the disease. X-ray mammography is the gold standard for diagnosing early signs of breast cancer, and Artificial Intelligence enables the detection of suspicious lesions and classifying them in terms of malignancy. This paper aimed to investigate mass detection and classification in a large-scale OPTIMAM dataset with 6000 cases and extracted 3524 images with masses in the mammograms of the Hologic manufacturer. The methodology of the detection step is to train the RetinaNet architecture of ResNet50, ResNet101, and ResNet152 backbones with three types of initializations by ImageNet and COCO weights and from scratch. The dataset was pre-processed to generate two types of input with entire mammograms and patches, which are stated as the first and the second approaches. The results show that in the first approach, RetinaNet of ResNet50 backbone with ImageNet and COCO weights and ResNet152 with the same weights performed 0.91 True Positive Rate at 0.78 False Positive Per Image, respectively. In contrast, in the second approach, ResNet152 with ImageNet weights reached 0.88 TPR at 0.78 FPPI. In the classification step, the Transfer Learning approach was applied with fine-tuning by adding L2-regularization and class weights to balance class distribution in the datasets
- …
