Angular Margins Research Articles

Integrating Mixup and ArcFace for enhanced anomalous sound detection

In machine condition monitoring, accurate detection of anomalous sounds is essential for timely maintenance interventions. Traditional deep neural network models employ classification-based self-supervised learning to discriminate normal and abnormal operational sounds using classification confidence. However, these models often suffer from overfitting or excessive confidence, limiting their effectiveness. To overcome this, we introduce a hybrid technique that combines the strengths of Mixup and ArcFace techniques. Mixup enhances model generalization by training on mixed sound data, while ArcFace promotes greater inter-class distances with an angular margin loss. However, there is a challenge in combining these methods because the angular margin for ArcFace is defined for a single target class, and hence, can be hardly defined for the mixed data utilized in Mixup. The proposed technique employs the class-dependent angular margins based on the mixing ratio of data. SoftMax losses for individual classes are constructed using the class-dependent margins and then integrated into the Mixup loss. When applied to the DCASE2020 Task2 dataset, the proposed method significantly outperforms the state-of-the-art NoisyArcMix model in anomaly detection. This approach demonstrates the potential of leveraging class-dependent angular margins in mixed data for improved performance in machine condition monitoring.

Comparative Analysis of the Diagnostic Value of S-Detect Technology in Different Planes Versus the BI-RADS Classification for Breast Lesions

Rationale and ObjectivesS-Detect, a deep learning-based Computer-Aided Detection system, is recognized as an important tool for diagnosing breast lesions using ultrasound imaging.However,it may exhibit inconsistent findings across multiple imaging planes. This study aims to evaluate the diagnostic performance of S-Detect in different planes and identify factors contributing to these inconsistencies. Materials and MethodsA retrospective cohort study was conducted on 711 patients with 756 breast lesions between January 2019 and January 2022. S-Detect was utilized to assess lesions in radial and anti-radial planes. BI-RADS classifications were employed for comparative analysis.The diagnostic performance was compared within each group, and p-values were computed for intergroup comparisons. Univariable and multivariable analyses were conducted to identify factors contributing to diagnostic inconsistency in S-Detect across planes. ResultsAmong 756 breast lesions, 668 (88.4%) exhibited consistent S-Detect outcomes across planes while 88 (11.6%) were inconsistent. In the consistent group, the diagnostic accuracy and area under the curve (AUC) of S-Detect were significantly higher than those of BI-RADS (accuracy: 91.2% vs. 84.9%, p = 0.045; AUC: 0.916 vs. 0.859, p = 0.036). In the inconsistent group, the diagnostic accuracy and AUC of S-Detect in radial and anti-radial planes were lower than those of BI-RADS (accuracy: 47.7% for radial, 52.2% for anti-radial vs. 69.3% for BI-RADS, p = 0.014, p-anti = 0.039; AUC: 0.503 for radial, 0.497 for anti-radial vs. 0.739 for BI-RADS, p = 0.042, p-anti < 0.001). Diagnostic inconsistency in S-Detect across planes was significantly associated with lesion size, indistinct or angular margins, and enhancement posterior acoustic features(p< 0.05). ConclusionS-Detect has outperformed BI-RADS in diagnostic precision under conditions of inter-planar concordance. However, its diagnostic efficacy is compromised in scenarios of inter-planar discordance. Under these circumstances, the results of S-Detect should be carefully referenced.

Evaluating the Efficacy of 6-Month Follow-Up for BI-RADS 3 Lesions Identified by Screening Ultrasound: A Retrospective Study

Background: The Breast Imaging-Reporting and Data System (BI-RADS), developed to standardize mammographic findings, categorizes 'probably benign' lesions as BI-RADS 3, aiming to reduce unnecessary biopsies from false positives, with a cancer likelihood of less than 2%. In some countries, screening programs have been expanded to include ultrasound alongside mammography, particularly in populations with predominantly dense breast tissue, leading to an increased detection of BI-RADS 3 lesions through this additional modality. This study examines the effectiveness of 6-month follow-ups for these ultrasound-identified BI-RADS 3 lesions. The purpose is to evaluate cancer detection rates and diagnostic performance of ultrasound findings leading to decision for biopsy. Materials and Methods: We retrospectively analyzed 7,577 women who underwent mammography and ultrasound screening from January 2016 to December 2021. Of these, 2,907 were classified as BI-RADS 3 based on ultrasound findings. The study focused on 1,163 patients with normal mammography and assessed BI-RADS 3 from ultrasound findings, who completed a 24-month follow-up. Data analysis included assessing demographic data, breast cancer risk, imaging features, and pathological findings. Results: Among the 1,163 patients, the cancer detection of the entire cohort was 0.7%, with no cancers detected during the first six months. The cancer found at 6 months and 12-24 months were 0 and 0.17%, respectively (p&lt;0.001). The median time to be reassesed from BI-RADS 3 to BI-RADS 4 was 18.4 months. Upon the change of BI-RADS, spiculation and angular margins were the most predictive ultrasound features for malignancy (AUROC=0.75 and 0.69, respectively). Lesion size growth alone was found insufficient as a biopsy criterion. A 28% growth cutoff distinguished between benign and malignant lesions better than 20% cutoff (AUROC=0.68 vs 0.64, p=0.038). Conclusion: A 12-month follow-up interval may be more appropriate than the traditional 6-month interval for average-risk patients with BI-RADS 3 lesions detected by screening ultrasound. Combining suspicious imaging features with size increases enhances diagnostic accuracy, proposing a tailored follow-up approach based on individual risk profiles and imaging characteristics.

Imaging Patterns of Breast Cancer in Women of Reproductive Age

Background: While breast cancer is less common in younger women compared to older age groups, it can still occur, necessitating careful evaluation of imaging findings. Imaging patterns of breast cancer in women of reproductive age can vary, presenting unique challenges in diagnosis and management. This study aimed to assess the imaging patterns of breast cancer in women of reproductive age. Methods: This prospective observational study was conducted in the Department of Radiology and Imaging Uttara Adhunik Medical College &amp; Hospital and Dhaka Medical College Hospital, Dhaka, Bangladesh from January 2019 to December 2023. In this study, 67 women of reproductive age who had been diagnosed with breast carcinoma and had undergone preoperative breast imaging were enrolled purposively. Data analysis was conducted using MS Office tools. Results: Mammographic findings: 42% had heterogeneously dense breasts, 30% scattered composition; and 87% had high mass density. Irregular mass shape: 43% (mammography), 69% (ultrasonography). In total 85% of cases were pleomorphic. Ultrasonography evaluation: 69% had a single US background. Echogenicity distribution: 61% hypoechoic among patients. MRI findings: 37% mild, 31% moderate, 22% minimal. Besides, 87% showed heterogeneous mass enhancement; 63% had segmental non-mass distribution. Conclusion: In breast cancer cases, imaging findings often include masses exhibiting suspicious features such as irregular shape and spiculated margins on mammograms, and irregular shape with angular margins on ultrasound. MRI features commonly involve masses with irregular shapes and heterogeneous enhancement.

X2-Softmax: Margin adaptive loss function for face recognition

Learning the discriminative features of different faces is an important task in face recognition. By extracting face features with neural networks, it becomes easy to measure the similarity of different face images, which makes face recognition possible. To enhance a neural network’s face feature separability, incorporating an angular margin during training is common practice. The state-of-the-art loss functions CosFace and ArcFace apply fixed margins between the weights of classes to enhance the inter-class separation of face features. Since the distribution of samples in the training set is uneven, the similarities between different identities are unequal. Therefore, using an inappropriately fixed angular margin may lead to problems such as that the model has difficulty converging or that the face features are not sufficiently discriminative. It is more intuitive to use adaptive angular margins are angular adaptive, which can increase as the angles between classes increase. In this paper, we propose a new angular margin loss named X2-Softmax. X2-Softmax loss has adaptive angular margins, which increase as the angle between different classes increases. The angular adaptive margin ensures model flexibility and effectively improves the effect of face recognition. We trained a neural network with X2-Softmax loss on the MS1Mv3 dataset and tested it on several evaluation benchmarks to demonstrate the effectiveness and superiority of our loss function.

Imaging Patterns in Breast Cancer for Women Under 40 Years: A Descriptive Cohort Study

Background and AimBreast cancer is the most frequently occurring malignant disease in women and remains the leading cause of cancer-related deaths among females worldwide. The aim of this study is to evaluate the imaging findings of breast cancer in women under the age of 40 and analyze their pathological patterns.MethodA retrospective study was conducted from 2013 to 2019, involving 120 patients below 40 years of age with pathologically confirmed primary epithelial breast cancers. The data were collected from the electronic records of a tertiary hospital in Riyadh, Saudi Arabia. Mammograms were performed for 115 patients, ultrasounds were conducted for all patients, and MRI scans were carried out for 47 patients.ResultsAll radiological findings and clinical characteristics of the 120 cases were retrieved from our digital-based system. The majority of breast cancer patients (83.4%) were between 30 and 40 years old, and the most common clinical presentation was a mass (45.8%). Out of the 73 patients who underwent genetic tests, 32.9% tested positive for gene mutations. No statistically significant correlation was found between specific age groups and breast composition (P = 0.216), specific mammogram abnormalities such as masses (P = 0.262), or microcalcifications (P = 0.421). Ultrasonography was performed for all patients, with abnormalities detected in only one patient who was diagnosed with Paget’s disease of the nipple. Masses, with or without parenchymal changes, were the predominant feature in 88.3% of cases.ConclusionThe imaging findings in breast cancer cases typically involve masses with suspicious features, irregular shape, and spiculated margins on mammograms, and irregular shape with microlobulated or angular margins on ultrasound. MRI features commonly include masses with irregular shape and heterogeneous enhancement. The luminal B subtype was identified as the most prevalent pathological feature, characterized by a high proliferative index (Ki-67%).Graphical

A Graph Convolutional Multiple Instance Learning on a Hypersphere Manifold Approach for Diagnosing Chronic Obstructive Pulmonary Disease in CT Images.

Chronic obstructive pulmonary disease (COPD) is a prevalent chronic disease with high morbidity and mortality. The early diagnosis of COPD is vital for clinical treatment, which helps patients to have a better quality of life. Because COPD can be ascribed to chronic bronchitis and emphysema, lesions in a computed tomography (CT) image can present anywhere inside the lung with different types, shapes and sizes. Multiple instance learning (MIL) is an effective tool for solving COPD discrimination. In this study, a novel graph convolutional MIL with the adaptive additive margin loss (GCMIL-AAMS) approach is proposed to diagnose COPD by CT. Specifically, for those early stage patients, the selected instance-level features can be more discriminative if they were learned by our proposed graph convolution and pooling with self-attention mechanism. The AAMS loss can utilize the information of COPD severity on a hypersphere manifold by adaptively setting the angular margins to improve the performance, as the severity can be quantified as four grades by pulmonary function test. The results show that our proposed GCMIL-AAMS method provides superior discrimination and generalization abilities in COPD discrimination, with areas under a receiver operating characteristic curve (AUCs) of 0.960 ± 0.014 and 0.862 ± 0.010 in the test set and external testing set, respectively, in 5-fold stratified cross validation; moreover, it demonstrates that graph learning is applicable to MIL and suggests that MIL may be adaptable to graph learning.

Sonografik Olarak Kuşkulu Olan ve Histopatolojik Olarak Benign Tanı Alan 1 cm’den Küçük Solid Meme Lezyonları: Sonografik Karakterizasyon

Aim: The aim of our study was to reveal the types and sonographic features of the margins in solid lesions less than 10 mm in dimension, considered suspicious for malignancy in breast ultrasonography, and histopathologically diagnosed as benign; and therefore, to recall the features that will facilitate the evaluation of radiology-pathology compatibility after biopsy. &#x0D; Material and Methods: This study was conducted with 82 women, with BI-RADS 4-5 lesions sonographically, between 2017 and 2020. Lesion size and margins, presence of posterior shadowing and microcalcifications were scanned retrospectively. Lesions were classified according to their margins as smooth-macrolobulated, microlobulated, irregular-indistinct, angular and spiculated. &#x0D; Results: Histopathologically, the most common benign lesions were fibroadenoma (n=26, 31.7%) and fibrocystic changes (n=15, 18.3%). Sonographically, the mean size of the lesions was 8.96±1.46 mm, and the most common margins were irregular-indistinct in 39%, and smooth-macrolobulated in 30%. In the statistical analysis, the incidence of fibroadenoma was found to be significantly higher in the BI-RADS 4a group compared to the patients in the other pathological diagnosis group (p:0.007). &#x0D; Conclusion: In this study, it was concluded that the indistinct-irregular, microlobulated and angular margins could also be observed significantly in subcentimeter benign breast lesions, and as the size of the lesion got smaller, it becomes difficult to differentiate the features of the margins; hence they should be evaluated more carefully.

RVFace: Reliable Vector Guided Softmax Loss for Face Recognition.

Face recognition has witnessed significant progress with the advances of deep convolutional neural networks (CNNs), and the central task of which is how to improve the feature discrimination. To this end, several margin-based (e.g., angular, additive and additive angular margins) softmax loss functions have been proposed to increase the feature margin between different classes. However, despite great achievements have been made, they mainly suffer from four issues: 1) They are based on the assumption of well-cleaned training sets, without considering the consequence of noisy labels inherently existing in most of face recognition datasets; 2) They ignore the importance of informative (e.g., semi-hard) features mining for discriminative learning; 3) They encourage the feature margin only from the perspective of ground truth class, without realizing the discriminability from other non-ground truth classes; and 4) They set the feature margin between different classes to be same and fixed, which may not adapt the situation of unbalanced data in different classes very well. To cope with these issues, this paper develops a novel loss function, which explicitly estimates the noisy labels to drop them and adaptively emphasizes the semi-hard feature vectors from the remaining reliable ones to guide the discriminative feature learning. Thus we can address all the above issues and achieve more discriminative features for face recognition. To the best of our knowledge, this is the first attempt to inherit the advantages of feature-based noisy labels detection, feature mining and feature margin into a unified loss function. Extensive experimental results on a variety of face recognition benchmarks have demonstrated the effectiveness of our method over state-of-the-art alternatives. Our source code is available at http://www.cbsr.ia.ac.cn/users/xiaobowang/.

CT Morphologic Characteristics and Variant Patterns of Interstitial Pulmonary Fibrosis in Systemic Lupus Erythematosus.

To assess CT features of pulmonary fibrosis in patients with systemic lupus erythematosus (SLE) and to assess the presence of several distinctive patterns of fibrosis associated with connective tissue disease. A cross-sectional retrospective analysis was performed. An institutional clinical database was queried for the years of 2005-2015 to identify CT examination reports of patients with SLE and fibrotic lung disease, which yielded 50 patients (median age, 49 years; age range, 22-71 years; 46 women). CT examination reports were scored by two subspecialty thoracic radiologists using a standard multilevel semiquantitative system. Readers noted the presence or absence of several recently described CT signs of variant patterns of fibrosis in connective tissue disease (the "anterior upper lobe," "straight-edge," and "exuberant honeycombing" signs), as well as two other morphologic characteristics (an "island-like" appearance of areas of well-defined fibrosis with angular margins surrounded by normal lung and confluent regions of lucent lung destruction). The most common CT patterns were characterized as either fibrotic nonspecific interstitial pneumonia (38%, 19 of 50) or variant fibrosis (44%, 22 of 50). CT signs of variant fibrosis were identified by both readers in up to 62% of patients, with good κ agreement (0.44-0.64); the island-like sign (62%) and anterior upper lobe sign (52%) were most commonly observed. Pulmonary function test results showed correlations with several imaging findings but did not show correlations with CT signs of variant fibrosis. When present, pulmonary fibrosis in SLE often has a distinctive appearance and may also manifest as several variant fibrotic patterns.Keywords: CT, Lung© RSNA, 2021See also the commentary by White in this issue.

BI-RADS Ultrasound Lexicon Descriptors and Stromal Tumor-Infiltrating Lymphocytes in Triple-Negative Breast Cancer

Increased levels of stromal tumor-infiltrating lymphocytes (sTILs) have recently been considered a favorable independent prognostic and predictive biomarker in triple-negative breast cancer (TNBC). The purpose of this study was to determine the relationship between BI-RADS (Breast Imaging Reporting and Data System) ultrasound lexicon descriptors and sTILs in TNBC. Patients with stage I-III TNBC were evaluated within a single-institution neoadjuvant clinical trial. Two fellowship-trained breast radiologists used the BI-RADS ultrasound lexicon to assess pretreatment tumor shape, margin, echo pattern, orientation, posterior features, and vascularity. sTILs were defined as low <20 or high ≥20 on the pretreatment biopsy. Fisher's exact tests were used to assess the association between lexicon descriptors and sTIL levels. The 284 patients (mean age 52 years, range 24-79 years) were comprised of 68% (193/284) with low-sTIL tumors and 32% (91/284) with high-sTIL tumors. TNBC tumors with high sTILs were more likely to have the following features: (1) oval/round shape than irregular shape (p = 0.003), (2) circumscribed or microlobulated margins than spiculated, indistinct, or angular margins (p = 0.0005); (3) complex cystic and solid pattern than heterogeneous pattern (p = 0.006); and (4) posterior enhancement than shadowing (p = 0.002). There was no significant association between sTILs and descriptors for orientation and vascularity (p = 0.06 and p = 0.49, respectively). BI-RADS ultrasound descriptors of the pretreatment appearance of a TNBC tumor can be useful in discriminating between tumors with low and high sTIL levels. Therefore, there is a potential use of ultrasound tumor characteristics to complement sTILs when used as stratification factors in treatment algorithms for TNBC.

English

BACKGROUND Lobular breast carcinomas have always been a diagnostic challenge, over the years, to the radiologist. They are one of the most commonly missed lesions on breast cancer screening checks, due to their varied presentation. We wanted to provide a concise and practical approach to characterise their morphology and presentation on mammography and ultrasound. METHODS A retrospective study was done for four years and a total of 699 patients with histopathologically proven breast cancer were chosen. Those patients with invasive lobular carcinoma (N = 56) and invasive ductal carcinoma (N = 538) were segregated and 50 cases from each group were selected randomly. RESULTS On mammography, an irregular, high-density mass was the most common presentation of both lobular (68 %) and ductal (86 %) carcinomas. Presentation as focal asymmetry (28 %) was significantly more prevalent in lobular breast carcinomas. Sonographically, an architectural distortion (30 %) and non-parallel orientation (28 %) was predominantly seen in invasive lobular carcinomas (ILCs). Mass (88 %) with micro lobulated (34 %) or angular (22 %) margins was more in favour of ductal carcinoma. Other general parameters like age at presentation, positive family history, multifocality, bilaterality, tumour size and lymph nodal involvement were not significantly different between both the groups. CONCLUSIONS A careful analysis of digital breast tomosynthesis and ultrasonography, keeping in mind all the clear differentiating features, along with experience in the field, will dramatically increase the early detection of lobular breast cancers. KEY WORDS Invasive Lobular Carcinoma (ILC), Invasive Ductal Carcinoma (IDC), X-Ray Mammography, Sono-Mammography

Bakhtiarellidae, a new end-Cretaceous ‘hauraniiform’ family of Larger Benthic Foraminifera: taxonomic inventory and phylogenetic assessment

ABSTRACT Bakhtiariella, a new genus of Larger Benthic Foraminifera is described from the upper Maastrichtian Tarbur Formation of Iran. The new taxon is characterised by a trochospirally coiled initial test part with angular chamber margins, later uncoiling and becoming rectilinear and presenting a coarse subepidermal network of horizontal and vertical partitions. Appearing as an extension of the subepidermal alveolar layer (exoskeleton), the pseudotriangular-shaped main partitions extend inwards and stretch downwards to the chamber floor where their arrangement is related to the inner margin of the central area with irregularly distributed pore-like foramina. Together with Persiacyclammina, which also displays such peculiar ‘subepidermal septula’, Bakhtiariella is included into the new family Bakhtiarellidae, and further compared with the Jurassic–Lower Cretaceous Hauraniidae. The new family gives further evidence for the diversification and pronounced endemism (Arabian Plate) of Larger Benthic Foraminifera within the final phase of the so-called Upper Cretaceous Global Community Maturation Cycle. This leads to the conclusion that both Persiacyclammina and Bakhtiariella can be considered ‘Key Mesozoic Benthic Foraminifera of the Middle East’.

Abstract PS3-16: Identifying the most relevant descriptors when evaluating ultrasound images in breast cancer diagnostics: A secondary analysis of an international multicenter trial

Abstract Background and objectives: The Breast Imaging Reporting and Data System (BI-RADS) provides a standardized way to describe ultrasound images in breast cancer diagnostics. However, there is little information which descriptors are most strongly associated with malignancy and to which extend the single descriptors (tissue composition, shape, orientation, margin of lesion, echo pattern, posterior features, and calcifications) should be considered for the final evaluation of risk of malignancy. Thus, we aimed to identify which BI-RADS descriptors are most strongly associated with malignancy when evaluating ultrasound images in breast cancer diagnostics. Methods: This multicenter, prospective trial took place at 11 trial sites in Austria, France, Germany, Japan, Netherlands, Portugal, and the US from February 2016 to March 2019. The trial enrolled 1288 women presenting with a lesion ≥0.5 and ≤5 cm in 2D B-mode ultrasound. The examiner conducted a routine 2D B-mode ultrasound examination and had additional standard information about the patients’ disease history and family history. The examiner described the ultrasound images according to BI-RADS. All patients underwent histopathological confirmation which was the gold standard against which the clinical examiner was compared. We performed univariate and multivariate analyses using descriptive statistics, Chi-Square test, and logistic regression to identify which image descriptors are associated with malignancy. Results: Histopathologic evaluation showed malignancy in 368 of 1288 lesions (28.6%). The descriptors most strongly associated with malignancy were spiculated margins (rate of malignancy 84.9%; 79 of 93), calcification (69.9%; 51 of 73), un-parallel orientation (65.9%; 187 of 284), angular margins (64.6%; 64 of 99), posterior shadowing (62.4%; 88 of 142), irregular shape (55.2%; 208 of 377), and indistinct margins (52.0%; 185 of 356). Different tissue compositions and echo patterns were least useful to distinguish between malign and benign lesions. Upon multivariate analysis, calcifications (OR 5.52; 95% CI 1.94-15.87) and posterior shadowing (OR 16.13; 95% CI 2.75-90.91) remained significantly (p&amp;lt;0.05) associated with malignancy. Conclusion: We identified which BI-RADS descriptors are most strongly associated with malignancy when evaluating ultrasound images in breast cancer diagnostics. Future research may look into providing not only a standardized image description but also a standardized final evaluation for the rate of malignancy with respect to the different predictive usefulness of the single descriptors. This may further standardize and objectify the risk evaluation in breast cancer diagnostics. Trial registration: NCT02638935 Table 1: Association of BI-RADS descriptors with final histopathologic resultsbenign pathologymalignant pathologyp-valuetissue compositionp&amp;lt;0.0001homogeneous background texture; fat —no. (%)185 (60.3)122 (39.74)homogeneous background texture; fibroglandular —no. (%)378 (77.3)111 (22.7)heterogeneous background texture —no. (%)356 (72.7)134 (27.4)shape of lesionp&amp;lt;0.0001oval —no. (%)659 (86.1)106 (13.8)round —no. (%)89 (62.7)53 (37.3)irregular —no. (%)169 (44.8)208 (55.2)orientation of lesionp&amp;lt;0.0001parallel —no. (%)806 (82.6)170 (17.42)not parallel —no. (%)97 (34.15)187 (65.9)margin of lesionp&amp;lt;0.0001circumcised —no. (%)644 (89.0)80 (11.0)indistinct margin —no. (%)171 (48.0)185 (52.0)angular margin —no. (%)35 (35.4)64 (64.6)microlobulated margin —no. (%)117 (60.0)78 (40.0)spiculated margin —no. (%)14 (15.1)79 (84.9)echo patternp=0.02anechoic —no. (%)7 (100)0 (0.0)hyperechoic —no. (%)30 (79.0)8 (21.0)complex cystic and solid —no. (%)52 (82.5)11 (17.5)hypoechoic —no. (%)645 (71.0)264 (29.0)isoechoic —no. (%)40 (78.4)11 (21.6)heterogeneous —no. (%)136 (64.8)74 (35.2)posterior featuresp&amp;lt;0.0001none —no. (%)590 (73.2)216 (26.8)enhancement —no. (%)249 (83.3)50 (16.7)shadowing —no. (%)53 (37.6)88 (62.4)combined pattern —no. (%)20 (58.8)14 (41.2)calcificationp&amp;lt;0.0001no calcification —no. (%)894 (73.8)317 (26.2)calcification —no. (%)22 (30.1)51 (69.9) Citation Format: André Pfob, Richard G. Barr, Volker Duda, Christopher Buesch, Joerg Heil, Michael Golatta. Identifying the most relevant descriptors when evaluating ultrasound images in breast cancer diagnostics: A secondary analysis of an international multicenter trial [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PS3-16.

Learning deep discriminative embeddings via joint rescaled features and log-probability centers

Recently softmax based loss functions have surged to advance image classification and face verification. Most efforts boost discrimination of the softmax loss by using novel angular margins in varying ways, but few analyze where the discrimination truly comes from whilst considering the power of relieving the overfitting to enhance the softmax loss. In this paper, we firstly delve into such mainstream of softmax based loss functions in theory, and recognize the importance of easing overfitting to the softmax loss. In terms of such analysis, this paper intends to bring the softmax loss up to the competitive level with current well-behaved loss functions. We do this in two ways: (1) regularizing the softmax to relieve the overfitting by learning the log-probability centers, and (2) rescaling deep embeddings of the softmax with a constant scale to further enhance inter-class separability in Euclidean space. We call the resulting loss function rLogCenter loss for short. Simple and interpretable as our loss is, it guides CNNs to induce performance gains in the experiments of both image classification and face verification.

Correlation between imaging features and molecular subtypes of breast cancer in young women (≤30years old).

To analyze the features of digital mammography (DM) plus digital breast tomosynthesis (DBT), ultrasonography (US) and magnetic resonance imaging (MRI) of breast cancer in young women (≤30years old) and the correlation with molecular subtypes. We performed a retrospective study of imaging features of consecutive young women aged ≤30years who were treated and surgically confirmed with breast cancer between January 2013 and December 2019 in our institution. All patients were Chinese women. DM + DBT and US were available for 170 lesions, MRI for 41 lesions. The imaging features were analysed by univariate and multivariate logistic regression analyses to find the predictive factors of the molecular subtypes. The predictive factors of the luminal B(HER2-) subtype (n = 51) were the mass with microcalcifications, irregular shape, spiculated margins, and shadowing posterior features (all P < 0.01). The predictive factors of the luminal B(HER2+) subtype (n = 26) were the spiculated margins (DBT + DM), angular margins (US), shadowing posterior features, and high vascularity (all P < 0.05). The predictive factors of the luminal A subtype (n = 37) were the mass without microcalcifications, spiculated margins, shadowing posterior features, and low vascularity (all P < 0.05). The predictive factors of the triple-negative subtype (n = 31) were the mass without microcalcifications, oval/round shape, circumscribed margins, enhancement of posterior features, and rim enhancement (MRI) (all P < 0.005). The predictive factors of the human-epidermal-growth-factor-receptor-2-enriched subtype (n = 26) were the only microcalcifications, microlobulated margins, and combined posterior feature (all P < 0.05). Compared with the general population of breast cancer, this young female population presents a different molecular phenotype distribution. Some imaging features of breast cancer in young women ≤30years old can be used to predict certain tumor molecular subtypes.

Mis-Classified Vector Guided Softmax Loss for Face Recognition

Face recognition has witnessed significant progress due to the advances of deep convolutional neural networks (CNNs), the central task of which is how to improve the feature discrimination. To this end, several margin-based (e.g., angular, additive and additive angular margins) softmax loss functions have been proposed to increase the feature margin between different classes. However, despite great achievements have been made, they mainly suffer from three issues: 1) Obviously, they ignore the importance of informative features mining for discriminative learning; 2) They encourage the feature margin only from the ground truth class, without realizing the discriminability from other non-ground truth classes; 3) The feature margin between different classes is set to be same and fixed, which may not adapt the situations very well. To cope with these issues, this paper develops a novel loss function, which adaptively emphasizes the mis-classified feature vectors to guide the discriminative feature learning. Thus we can address all the above issues and achieve more discriminative face features. To the best of our knowledge, this is the first attempt to inherit the advantages of feature margin and feature mining into a unified loss function. Experimental results on several benchmarks have demonstrated the effectiveness of our method over state-of-the-art alternatives. Our code is available at http://www.cbsr.ia.ac.cn/users/xiaobowang/.

Canine Bite-Mark Evidence in Veterinary Necropsy: Case Studies Featuring the Bite-Mark Examination

Forensic veterinary medicine tends to gain increasing importance in veterinary pathology, the number of such cases being still limited. The paper describes cases with multiple cutaneous wounds, with emphasis on bite-mark analysis. The material was represented by corpses (deer and dogs) brought by local authorities to Faculty of Veterinary Medicine Cluj-Napoca (Romania), along with preliminary reports issued by the competent organs. A complete medicolegal survey was done. The cases analyzed by us displayed numerous usually symmetrical (opposed) wounds, with angular margins and elongated aspect (somehow specific to bite wounds). All cases assessed presented laceration of the skin from the subcutaneous tissue, rupture of muscle and (in some cases) perforation of internal organs. The features of the bite-marks described were typical for carnivores. Depending on the species involved, the topography of lesions varied. In deer, the ventrolateral area of the neck and the lumbar/dorsal region were more commonly affected. The distribution suggests the hunting instinct of carnivores by trying to induce suffocation of the prey. In the analyzed dog corpses, the most affected areas were the auricular, limbs, sternal region, abdominal region. Investigation of bite-marks is an evolving field of forensic veterinary science and remains a provocation for veterinary pathologists.

Deep Imbalanced Learning for Face Recognition and Attribute Prediction.

Data for face analysis often exhibit highly-skewed class distribution, i.e., most data belong to a few majority classes, while the minority classes only contain a scarce amount of instances. To mitigate this issue, contemporary deep learning methods typically follow classic strategies such as class re-sampling or cost-sensitive training. In this paper, we conduct extensive and systematic experiments to validate the effectiveness of these classic schemes for representation learning on class-imbalanced data. We further demonstrate that more discriminative deep representation can be learned by enforcing a deep network to maintain inter-cluster margins both within and between classes. This tight constraint effectively reduces the class imbalance inherent in the local data neighborhood, thus carving much more balanced class boundaries locally. We show that it is easy to deploy angular margins between the cluster distributions on a hypersphere manifold. Such learned Cluster-based Large Margin Local Embedding (CLMLE), when combined with a simple k-nearest cluster algorithm, shows significant improvements in accuracy over existing methods on both face recognition and face attribute prediction tasks that exhibit imbalanced class distribution.

EGFR-TKI-sensitive mutations in lung carcinomas: are they related to clinical features and CT findings?

BackgroundEpidermal growth factor receptor (EGFR) mutation testing is restricted to several limitations. In this study, we examined the relationship between EGFR mutation status and clinicoradiological characteristics in a Chinese cohort of patients.Materials and methodsThe data of patients who were diagnosed with lung carcinoma and underwent both EGFR testing and chest computed tomography (CT) at our hospital between January 1, 2011, and November 31, 2015, were retrospectively analyzed. The age, sex, and smoking index of the patients, the size, margin, and density of the tumor, and the presence of specific signs visible on the CT images were assessed.ResultsThe results showed a higher rate of EGFR-tyrosine kinase inhibitor (TKI)-sensitive group than nonsensitive group in female patients and patients with a low smoking index (P<0.001, both). In logistic regression analyses, tumor size (P<0.001), smooth margins (P=0.015), and angular margins (P<0.001) were independent negative predictors of EGFR-TKI-sensitive group. Pleural indentation (P<0.001) and air bronchogram (P=0.025) were independent positive predictors of EGFR-TKI-sensitive group. Patients with squamous cell carcinoma had fewer sensitive mutations than those with either adenocarcinoma (P<0.001) or adenosquamous carcinoma (P<0.001).ConclusionClinical and CT characteristics differed significantly between EGFR-TKI-sensitive and nonsensitive groups. Our findings may be useful in deciding therapeutic strategies for patients in whom EGFR testing is not possible.