Ziba Gandomkar

Computer-based image analysis in breast pathology

Whole slide imaging (WSI) has the potential to be utilized in telepathology, teleconsultation, quality assurance, clinical education, and digital image analysis to aid pathologists. In this paper, the potential added benefits of computer-assisted image analysis in breast pathology are reviewed and discussed. One of the major advantages of WSI systems is the possibility of doing computer-based image analysis on the digital slides. The purpose of computer-assisted analysis of breast virtual slides can be (i) segmentation of desired regions or objects such as diagnostically relevant areas, epithelial nuclei, lymphocyte cells, tubules, and mitotic figures, (ii) classification of breast slides based on breast cancer (BCa) grades, the invasive potential of tumors, or cancer subtypes, (iii) prognosis of BCa, or (iv) immunohistochemical quantification. While encouraging results have been achieved in this area, further progress is still required to make computer-based image analysis of breast virtual slides acceptable for clinical practice.

A model based on temporal dynamics of fixations for distinguishing expert radiologists' scanpaths

This study investigated a model which distinguishes expert radiologists from less experienced radiologists based on features describing spatio-temporal dynamics of their eye movement during interpretation of digital mammograms. Eye movements of four expert and four less experienced radiologists were recorded during interpretation of 120 two-view digital mammograms of which 59 had biopsy proven cancers. For each scanpath, a two-dimensional recurrence plot, which represents the radiologist’s refixation pattern, was generated. From each plot, six features indicating the spatio-temporal dynamics of fixations were extracted. The first feature measured the percentage of recurrent fixations; the second indicated the percentage of recurrent fixations which was fixated later in several consecutive fixations; the third measured the percentage of recurrent fixations that form a repeated sequence of fixations and the fourth assessed whether the recurrent fixations were occurring sequentially close together. The number of switches between the two mammographic views was also measured, as was the average number of consecutive fixations in each view before switching. These six features along with total time on case and average fixation duration were fed into a support vector machine whose performance was evaluated using 10-fold cross validation. The model achieved a sensitivity of 86.3% and a specificity of 85.2% for distinguishing experts’ scanpaths. The obtained result suggests that spatio-temporal dynamics of eye movements can characterize expertise level and has potential applications for monitoring the development of expertise among radiologists as a result of different training regimes and continuing education schemes.

iCAP: An Individualized Model Combining Gaze Parameters and Image-Based Features to Predict Radiologists’ Decisions While Reading Mammograms

This study introduces an individualized tool for identifying mammogram interpretation errors, called eye-Computer Assisted Perception (iCAP). iCAP consists of two modules, one which processes areas marked by radiologists as suspicious for cancer and classifies these as False Positive (FP) or True Positive (TP) decisions, while the second module classifies fixated but not marked locations as False Negative (FN) or True-Negative (TN) decisions. iCAP relies on both radiologists’ gaze-related parameters, extracted from eye tracking data, and image-based features. In order to evaluate iCAP, eye tracking data from eight breast radiologists reading 120 two-view digital mammograms were collected. Fifty-nine cases had biopsy proven cancer. For each radiologist, a user-specific support vector machine model was built to classify the radiologist’ s reported areas as TPs or FPs and fixated locations as TNs or FNs. The performances of the classifiers were evaluated by utilizing leave-one-out cross validation. iCAP was tested retrospectively in a simulated scenario in which it was assumed that the radiologists would accept all iCAP decisions. Using iCAP led to an average increase of 12%±6% in the number of correctly localized cancer and an average decrease of 44.5%±22.7% in the number of FPs per image.

Recurrence quantification analysis of radiologists' scanpaths when interpreting mammograms

Purpose The purpose of this study was to Propose a classifier based on recurrence quantification analysis (RQA) metrics for distinguishing experts’ scanpaths from those of less‐experienced readers and to explore the association of spatiotemporal dynamics of the mammographic scanpaths with the characteristics of cases and radiologists using RQA metrics. Materials and methods Eye movements were recorded from eight radiologists (two cohorts: four experienced and four less‐experienced) while reading 120 mammograms (59 cancer, 61 normal). Ten RQA measures were extracted for each recorded scanpath. The measures described the temporal distribution of recurrent fixations as well as laminar and deterministic eye movements. Recurrent fixations are fixations that are located close to a previously fixated point in a scanpath. Deterministic eye movements represent looking back and forth between two locations, while laminar eye movements indicate detailed scanning of an area with consecutive fixations. The RQA metrics along with six conventional eye‐tracking parameters were used to construct a classifier for distinguishing experts’ scanpaths from those of less‐experienced readers. Leave‐one‐out cross validation was used for evaluating the classifier. For each reader cohort, the ANOVA analysis was done to study the relationship of RQA measures with breast density, case pathology, readers’ expertise, and readers’ decisions on the case. The proportions of laminar and deterministic movements involved fixations in the location of lesions were also compared for two reader cohorts using two proportion z‐tests. Results All RQA measures differed significantly between scanpaths of experienced readers and those of less‐experienced readers. The classifier achieved an area under the receiver operating characteristic curve of 0.89 (0.87–0.91) for detecting experts’ scanpaths. Proportionately more refixations and laminar and deterministic sequences were in the location of the lesion for the experienced cohort compared to the less‐experienced cohort (all P‐values < 0.001). Eight and four RQA measures differed between normal and cancer cases for the experienced and less experienced readers, respectively. None of metrics differed between fatty and dense breasts for the less experienced readers, while two measures resulted into a significant difference for the experienced readers. For experts, six measures differed significantly between true negatives and false positives and nine were significantly different between true positives and false negatives. For the less‐experienced cohort, the corresponding figures were seven and one measures, respectively. Conclusion The RQA measures can quantify the differences among experienced and less experienced radiologists. They also capture differences among experts’ scanpaths related to case pathology and radiologists’ decisions on the case.

Predicting radiologists' true and false positive decisions in reading mammograms by using gaze parameters and image-based features

Radiologists’ gaze-related parameters combined with image-based features were utilized to classify suspicious mammographic areas ultimately scored as True Positives (TP) and False Positives (FP). Eight breast radiologists read 120 two-view digital mammograms of which 59 had biopsy proven cancer. Eye tracking data was collected and nearby fixations were clustered together. Suspicious areas on mammograms were independently identified based on thresholding an intensity saliency map followed by automatic segmentation and pruning steps. For each radiologist reported area, radiologist’s fixation clusters in the area, as well as neighboring suspicious areas within 2.5° of the center of fixation, were found. A 45-dimensional feature vector containing gaze parameters of the corresponding cluster along with image-based characteristics was constructed. Gaze parameters included total number of fixations in the cluster, dwell time, time to hit the cluster for the first time, maximum number of consecutive fixations, and saccade magnitude of the first fixation in the cluster. Image-based features consisted of intensity, shape, and texture descriptors extracted from the region around the suspicious area, its surrounding tissue, and the entire breast. For each radiologist, a userspecific Support Vector Machine (SVM) model was built to classify the reported areas as TPs or FPs. Leave-one-out cross validation was utilized to avoid over-fitting. A feature selection step was embedded in the SVM training procedure by allowing radial basis function kernels to have 45 scaling factors. The proposed method was compared with the radiologists’ performance using the jackknife alternative free-response receiver operating characteristic (JAFROC). The JAFROC figure of merit increased significantly for six radiologists.

Radiologists can detect the ‘gist’ of breast cancer before any overt signs of cancer appear

Radiologists can detect abnormality in mammograms at above-chance levels after a momentary glimpse of an image. The study investigated this instantaneous perception of an abnormality, known as a “gist” response, when 23 radiologists viewed prior mammograms of women that were reported as normal, but later diagnosed with breast cancer at subsequent screening. Five categories of cases were included: current cancer-containing mammograms, current mammograms of the normal breast contralateral to the cancer, prior mammograms of normal cases, prior mammograms with visible cancer signs in a breast from women who were initially reported as normal, but later diagnosed with breast cancer at subsequent screening in the same breast, and prior mammograms without any visible cancer signs from women labelled as initially normal but subsequently diagnosed with cancer. Our findings suggest that readers can distinguish patients who were diagnosed with cancer, from individuals without breast cancer (normal category), at above-chance levels based on a half-second glimpse of the mammogram even before any lesion becomes visible on the mammogram. Although 20 of the 23 radiologists demonstrated this ability, radiologists’ abilities for perceiving the gist of the abnormal varied between the readers and appeared to be linked to expertise. These results could have implications for identifying women of higher than average risk of a future malignancy event, thus impacting upon tailored screening strategies.

A cognitive approach to determine the benefits of pairing radiologists in mammogram reading.

Mammography screening in Europe and Australia is carried out by having two radiologists independently read the case and determine whether an actionable finding is present. If they disagree, a third radiologist – the arbitrator – reads the case and offers the final opinion. Currently radiologists are picked for the pair based on scheduling convenience, with no thought being given to whether a given pair of radiologists should really be put together to read cases. In the past research has shown that breast radiologists tend to commit the same mistakes time and again and incline to search mammograms in a particular way; hence, pairing two radiologists that tend to search a mammogram in an almost similar manner, for example, may not be such a good idea. In this study, we used eye position tracking to determine how radiologists searched a given set of cases. Using different cognitive models we paired the radiologists and determined the effect of the pairing on the radiologist’s performance using the Receivers Operating Characteristic Area Under the Curve (ROC AUC). Our results suggest that some pairings are detrimental to performance and should not be put together.

Mammographic density and other risk factors for breast cancer among women in China

Breast cancer is the most common neoplasm diagnosed among females in China and it is one of the leading causes of female cancer death, however the risk factors for breast cancer are not fully understood for Chinese women. One of the key risk factors shown to be relevant for westernized women is mammographic density but previously used observer Breast Imaging Reporting and Data System (BIRADS) technique to assess density is shown to have wide interand intraobserver variations. Therefore, quantitative techniques are increasingly recommended to assess this important parameter. The aim of the current study is to identify risk factors of breast cancer for Chinese women, with attention paid to mammographic density using quantitative measurements. This study was approved by the Human Research Ethics Committee of the University of Sydney (Project number: 2014/768). Women of 84 with and 987 without breast cancer were randomly selected from Fudan University Shanghai Cancer Center (FUSCC) from March 2015 to June 2016. The women with breast cancer were diagnosed within the hospital environment at FUSCC, while the other women were recruited from the Breast Cancer Screening Trail (BCST) organized by FUSCC. Demographic, lifestyle and reproductive characteristics were obtained from the registration form and the discharge summary in the health record for each woman with breast cancer and through a BCST questionnaire for breast cancerfree women. For all of the women, mammograms were acquired for cranio-caudal projection of both breasts. Mammographic density was measured by a fully automatic algorithm AutoDensity, which identifies both dense and breast areas in mammograms and then classifies mammographic density. Differences in characteristics between cancer and cancer-free women were assessed using t tests and chisquare tests. Binary logistic regression was then conducted for variables that were statistically significant from either the t test or the chi-squared test to produce odds ratios and 95% confidence intervals. Categorical variables with 0 frequency in any one of the categories were excluded from this test. The whole data set was then divided into two subsets based on menopause status, one for premenopausal and another for postmenopausal women. The statistical tests mentioned above were repeated for each subset. Table 1 shows the baseline differences of characteristics for two groups of women, and the outputs from binary logistic regression. Overall, it appears that large breast area, increasing age, increasing BMI, later age at menarche, earlier age at first delivery, longer duration of breastfeeding, postmenopause status, greater number of children, and a breastfeeding history are important agents. The results for pre and postmenopausal women are shown in Tables S1 and S2, respectively. The rest of this commentary however will focus on the implications around our findings on mammographic density. We failed to identify any association for mammographic density with breast cancer using percentage or dense area parameters, a finding which is consistent and inconsistent with previous work: one previous study which recruited 86 and 28 302 women with and without breast cancer, respectively, from a screening trial across 4 Chinese cities of similar size to our study also showed no association between density and cancer; in contrast another large cross-sectional study, involving 2527 cancer and 3394 cancer-free women, reported that, compared to women without breast cancer, mammographic density was lower and higher for cancer women within the 40-49 and 55-71 age groups, respectively, however there was no association for women aged 50-54. This difference between our work and the latter study might be explained by the fact that agedependent variations were not assessed in our work, thereby obscuring specific observations. Instead, we focused on categorizing our women based on menopausal status. Another possible Tong Li and Lichen Tang share the joint first authorship. Received: 9 January 2017 | Revised: 17 January 2017 | Accepted: 19 January 2017 DOI: 10.1111/tbj.12967

A framework for distinguishing benign from malignant breast histopathological images using deep residual networks.

Studies have shown that there are discrepancies among pathologists in the classification of breast histopathological slides. In this study we propose a framework for categorizing hematoxylin-eosin stained breast images either as benign or malignant at four magnification factors, and then aggregating the classification results of a patient’s images from different magnification factors to make the ultimate diagnosis for each patient. We used a publicly available database, containing 7786 images from 81 patients. The images were acquired in four visual magnification factors, namely x40, x100, x200, and x400, with an effective pixel size of 0.49 μm, 0.20 μm, 0.10 μm, and 0.05 μm respectively. In order to mitigate the inconsistencies in the color of the images, stain normalization was performed. Next, for each magnification factor, a deep residual network (ResNet) with 152 layers has been trained for classifying patches from the images as benign or malignant. Then, a meta-decision tree was used to combine classification results of a patient’s images from different magnification factors to provide a patient-level diagnosis. The ResNets achieved correct classification rates (CCR) of 98.52%, 97.90%, 98.33%, and 97.66% at x40, x100, x200, and x400 magnification factors, respectively. For classification of patients either as benign or malignant, a CCR of 98.77% was obtained. In conclusion, our study showed that the proposed framework can be helpful in the categorization of breast digital slides.

Detection of the abnormal gist in the prior mammograms even with no overt sign of breast cancer.

Can radiologists distinguish prior mammograms with no overt signs of cancer from women who were later diagnosed with breast cancer from the prior mammograms of women reported as normal and subsequently confirmed to be cancerfree? Twenty-three radiologists and breast physicians viewed 200 craniocaudial mammograms for a half-second and rated whether the woman would be recalled on a scale of 0 (clearly normal) to 100 (clearly abnormal). The dataset included five categories of mammograms, with each category containing 40 cases. The categories were Cancer (current cancer-containing mammograms), Prior-Vis (prior mammograms with visible cancer signs), Contra (current ‘normal’ mammograms contralateral to the cancer), Prior-Invis (priors without visible cancer signs), and Normal (priors of normal cases). For each radiologist, four pairs of analyses were performed to evaluate whether the radiologists could distinguish mammograms in each category from the normal mammograms: Cancer vs Normal, Prior-Vis vs Normal, Contra vs Normal, and Prior-Invis vs Normal. The Area under Receiver Operating Characteristic curves (AUC) was calculated for each paired grouping and each radiologist. Wilcoxon Signed Rank test showed the AUC values were above-chance for all comparisons: Cancer (z=4.20, P<0.001); Prior-Vis (z=4.11, P<0.001); Contra (z=4.17, P<0.001); Prior-Invis (z=3.71, P<0.001). The results suggest that radiologists can distinguish patients who were diagnosed with cancer from individuals without breast cancer at an above-chance level based on a half-second glimpse of mammogram even before the lesion becomes apparently visible (Prior-Invis). Apparently, something about the breast parenchyma can look abnormal before the appearance of a localized lesion