Application of Deep Convolution Network to Automated Image Segmentation of Chest CT for Patients With Tumor

ObjectivesTo automate image delineation of tissues and organs in oncological radiotherapy by combining the deep learning methods of fully convolutional network (FCN) and atrous convolution (AC).MethodsA total of 120 sets of chest CT images of patients were selected, on which radiologists had outlined the structures of normal organs. Of these 120 sets of images, 70 sets (8,512 axial slice images) were used as the training set, 30 sets (5,525 axial slice images) as the validation set, and 20 sets (3,602 axial slice images) as the test set. We selected 5 published FCN models and 1 published Unet model, and then combined FCN with AC algorithms to generate 3 improved deep convolutional networks, namely, dilation fully convolutional networks (D-FCN). The images in the training set were used to fine-tune and train the above 8 networks, respectively. The images in the validation set were used to validate the 8 networks in terms of the automated identification and delineation of organs, in order to obtain the optimal segmentation model of each network. Finally, the images of the test set were used to test the optimal segmentation models, and thus we evaluated the capability of each model of image segmentation by comparing their Dice coefficients between automated and physician delineation.ResultsAfter being fully tuned and trained with the images in the training set, all the networks in this study performed well in automated image segmentation. Among them, the improved D-FCN 4s network model yielded the best performance in automated segmentation in the testing experiment, with an global Dice of 87.11%, and a Dice of 87.11%, 97.22%, 97.16%, 89.92%, and 70.51% for left lung, right lung, pericardium, trachea, and esophagus, respectively.ConclusionWe proposed an improved D-FCN. Our results showed that this network model might effectively improve the accuracy of automated segmentation of the images in thoracic radiotherapy, and simultaneously perform automated segmentation of multiple targets.

High precision localization of pulmonary nodules on chest CT utilizing axial slice number labels

Background Reidentification of prior nodules for temporal comparison is an important but time-consuming step in lung cancer screening. We develop and evaluate an automated nodule detector that utilizes the axial-slice number of nodules found in radiology reports to generate high precision nodule predictions. Methods 888 CTs from Lung Nodule Analysis were used to train a 2-dimensional (2D) object detection neural network. A pipeline of 2D object detection, 3D unsupervised clustering, false positive reduction, and axial-slice numbers were used to generate nodule candidates. 47 CTs from the National Lung Cancer Screening Trial (NLST) were used for model evaluation. Results Our nodule detector achieved a precision of 0.962 at a recall of 0.573 on the NLST test set for any nodule. When adjusting for unintended nodule predictions, we achieved a precision of 0.931 at a recall 0.561, which corresponds to 0.06 false positives per CT. Error analysis revealed better detection of nodules with soft tissue attenuation compared to ground glass and undeterminable attenuation. Nodule margins, size, location, and patient demographics did not differ between correct and incorrect predictions. Conclusions Utilization of axial-slice numbers from radiology reports allowed for development of a lung nodule detector with a low false positive rate compared to prior feature-engineering and machine learning approaches. This high precision nodule detector can reduce time spent on reidentification of prior nodules during lung cancer screening and can rapidly develop new institutional datasets to explore novel applications of computer vision in lung cancer imaging.

A pipeline to characterize local cortical folds by mapping them to human-interpretable shapes

BackgroundVariations in regional cortical folds across individuals have been examined using computationally-derived morphological measures, or by manual characterization procedures that map distinct variants of a regional fold to a set of human-interpretable shapes. Although manual mapping approaches have proven useful for identifying morphological differences of clinical relevance, such procedures are subjective and not amenable to scaling.New MethodWe propose a 3-step pipeline to develop computational models of manual mapping. The steps are: represent regional folds as feature vectors, manually map each feature vector to a shape-variant that the underlying fold represents, and train classifiers to learn the mapping.ResultsFor demonstration, we chose a 2D-problem of detecting within slice discontinuity of medial and lateral sulci of orbitofrontal cortex (OFC); the discontinuity may be visualized as a broken H-shaped pattern, and is fundamental to OFC-type-characterization. The classifiers predicted discontinuities with 86-95% test-accuracy.Comparison with Existing MethodsThere is no existing pipeline that automates a manual characterization process. For the current demonstration problem, we conduct multiple analyses using existing softwares to explain our design decisions, and present guidelines for using the pipeline to examine other regional folds using conventional or non-conventional morphometric measures.ConclusionWe show that this pipeline can be useful for determining axial-slice discontinuity of sulci in the OFC and can learn structural-features that human-raters may rely on during manual-characterization.The pipeline can be used for examining other regional folds and may facilitate discovery of various statistically-reliable 2D or 3D human-interpretable shapes that are embedded throughout the brain.

Inter-rater reliability of TT-TG distance is good and does not vary based on preselected versus independent slice selection on MRI

ObjectivesPatellar instability is a frequent cause of knee dysfunction in young, active patients. Tibial tubercle–trochlear groove (TT-TG) distance, trochlear morphology (trochlear depth and sulcus angle) and patellar height are felt to contribute to patellar instability and may influence treatment. These measurements are frequently performed on MRI images. We hypothesised that inter-rater reliability of measures would be good and that inter-rater variation is driven primarily by slice selection.MethodsTwenty-six patients with at least one documented episode of patellar instability confirmed by MRI were identified. Six raters reviewed MRI images from each patient. Each rater measured and recorded TT-TG distance, trochlear depth and sulcus angle, and patellotrochlear index (PTI) for each patient and the slices used for the measurements. Each rater repeated the measurement using preselected slices. Inter-rater reliability was calculated by intraclass correlations (ICCs) for slice selection and for TT-TG distance, trochlear morphology measures and PTI with both independently selected and preselected slices. Statistically significant differences (p<0.05) in ICC based on slice selection were defined as values without overlap of their 95% CIs.ResultsInter-rater reliability was excellent for tibial (ICC=0.93) axial slice selection and sagittal slice selection (ICC=0.94), and good for femoral (ICC=0.88) axial slice selection. Using independent slice selection, inter-rater reliability was good for TT-TG distance (ICC=0.79) and fair for trochlear depth (ICC=0.57), sulcus angle (ICC=0.57) and PTI (ICC=0.71). When preselected slices were used, inter-rater reliability was good for TT-TG distance (ICC=0.85), sulcus angle (ICC=0.83) and PTI (ICC=0.77) and fair for trochlear depth (ICC=0.68). Only sulcus angle demonstrated a significant (p<0.05) improvement in inter-rater reliability with the use of preselected slices.Discussion and conclusionInter-rater reliability of TT-TG distance is good and does not vary based on preselected versus independent slice selection on MRI. Inter-rater reliability of trochlear morphology measures based on axial MRI slices and PTI is fair. Inter-rater variation can be reduced (particularly in the case of sulcus angle) through agreement on slice selection.Level of evidenceLevel III, diagnostic.

Correlation between subcutaneous fat measurements in knee MRI and BMI: relationship to obesity and related co-morbidities

Background Obesity is a major public health disorder associated with multiple co-morbidities. Knee magnetic resonance imaging (MRI) permits visualization of the subcutaneous fat anatomy, which can be correlated to body mass index (BMI) and obesity-related co-morbidities. Purpose This study intends to validate a method of correlating measurements of subcutaneous fat around the distal femur on axial MR images to BMI and obesity-related co-morbidities. Material and Methods The most proximal axial slice of each knee MRI was divided into four quadrants. Measurements of the thickest portion of the subcutaneous fat in each quadrant were independently obtained, yielding a value which was assigned the name of the SubCut fat index. The relationship between the SubCut fat index of each quadrant and the patient’s BMI was then evaluated. Receiver operating characteristic curves utilizing both the subcutaneous fat in the medial and lateral quadrants as well as BMI were performed with respect to obesity-related co-morbidities. Results SubCut fat index measurements in all four quadrants and BMI show the strongest correlation (all four, ANOVA P < 0.0001, r = 0.6), with subcutaneous fat measurements of the anterior medial (p < 0.0001) and posterior medial quadrants ( P = 0.01). Additionally, BMI and medial quadrants SubCut indices showed strong association with obesity-related co-morbidities including sleep apnea, asthma, diabetes, hypertension, gastroesophageal reflux disease, and osteoporosis. Conclusion The SubCut fat index, a marker of distal femur subcutaneous fat on axial MRI, correlates with severity of obesity (BI) and associated obesity-related co-morbidities.

Anterior centre-edge angle on sagittal CT: a comparison of normal hips to dysplastic hips

Purpose: The anterior centre-edge angle (ACEA) describes anterior acetabular coverage on false profile radiographs. Variability associated with pelvic tilt, radiographic projection, and identifying the true anterior edge, causes discrepancies in measuring an accurate ACEA. Computed tomography (CT) has the potential of improving the accuracy of ACEA. However, because the ACEA on sagittal CT has been shown to not be equivalent to ACEA on false profile radiographs, the normal range of ACEA on CT currently remains unknown and cannot reliably be used to determine over/under coverage. We therefore asked: what is the normal variation of ACEA corrected for pelvic tilt on sagittal CT and how does this compare to dysplastic hips? Material and Methods: A retrospective review was conducted on patients 10–35 who underwent CT for non-orthopedic related issues and patients with known hip dysplasia. The ACEA was measured on a sagittal slice corresponding to the centre of the femoral head on the axial slice and adjusted for pelvic tilt. A statistical comparison was then performed. Results: A total of 320 normal patients and 22 patients with hip dysplasia were reviewed. The mean ACEA for all ages was 50° ± 8°, (range: 23–81º), with a larger mean ACEA for males (51°) than females (49°). The ACEA mean for dysplastic hips was 30° ± 11° with a statistically significant difference in mean from the normal hip group ( p < 0.0001). Conclusion: The ACEA can be reliably measured on sagittal CT and significantly differs from dysplastic hips. ACEA measurements above 66° or below 34° may represent anterior over and under coverage.