User Interaction in Semi-Automatic Segmentation of Organs at Risk: a Case Study in Radiotherapy.

Accurate segmentation of organs at risk is an important step in radiotherapy planning. Manual segmentation being a tedious procedure and prone to inter- and intra-observer variability, there is a growing interest in automated segmentation methods. However, automatic methods frequently fail to provide satisfactory result, and post-processing corrections are often needed. Semi-automatic segmentation methods are designed to overcome these problems by combining physicians' expertise and computers' potential. This study evaluates two semi-automatic segmentation methods with different types of user interactions, named the "strokes" and the "contour", to provide insights into the role and impact of human-computer interaction. Two physicians participated in the experiment. In total, 42 case studies were carried out on five different types of organs at risk. For each case study, both the human-computer interaction process and quality of the segmentation results were measured subjectively and objectively. Furthermore, different measures of the process and the results were correlated. A total of 36 quantifiable and ten non-quantifiable correlations were identified for each type of interaction. Among those pairs of measures, 20 of the contour method and 22 of the strokes method were strongly or moderately correlated, either directly or inversely. Based on those correlated measures, it is concluded that: (1) in the design of semi-automatic segmentation methods, user interactions need to be less cognitively challenging; (2) based on the observed workflows and preferences of physicians, there is a need for flexibility in the interface design; (3) the correlated measures provide insights that can be used in improving user interaction design.

Use of the Resection Map system as guidance during hepatectomy.

BACKGROUND: The objective of this work is to evaluate a new concept of intraoperative three-dimensional (3D) visualization system to support hepatectomy. The Resection Map aims to provide accurate cartography for surgeons, who can therefore anticipate risks, increase their confidence and achieve safer liver resection. METHODS: In an experimental prospective cohort study, ten consecutive patients admitted for hepatectomy to three European hospitals were selected. Liver structures (portal veins, hepatic veins, tumours and parenchyma) were segmented from a recent computed tomography (CT) study of each patient. The surgeon planned the resection preoperatively and read the Resection Map as reference guidance during the procedure. Objective (amount of bleeding, tumour resection margin and operating time) and subjective parameters were retrieved after each case. RESULTS: Three different surgeons operated on seven patients with the navigation aid of the Resection Map. Veins displayed in the Resection Map were identified during the surgical procedure in 70.1% of cases, depending mainly on size. Surgeons were able to track resection progress and experienced improved orientation and increased confidence during the procedure. CONCLUSIONS: The Resection Map is a pragmatic solution to enhance the orientation and confidence of the surgeon. Further studies are needed to demonstrate improvement in patient safety.

Stacking denoising auto-encoders in a deep network to segment the brainstem on MRI in brain cancer patients: A clinical study.

Delineation of organs at risk (OARs) is a crucial step in surgical and treatment planning in brain cancer, where precise OARs volume delineation is required. However, this task is still often manually performed, which is time-consuming and prone to observer variability. To tackle these issues a deep learning approach based on stacking denoising auto-encoders has been proposed to segment the brainstem on magnetic resonance images in brain cancer context. Additionally to classical features used in machine learning to segment brain structures, two new features are suggested. Four experts participated in this study by segmenting the brainstem on 9 patients who underwent radiosurgery. Analysis of variance on shape and volume similarity metrics indicated that there were significant differences (p<0.05) between the groups of manual annotations and automatic segmentations. Experimental evaluation also showed an overlapping higher than 90% with respect to the ground truth. These results are comparable, and often higher, to those of the state of the art segmentation methods but with a considerably reduction of the segmentation time.

Supervised machine learning-based classification scheme to segment the brainstem on MRI in multicenter brain tumor treatment context.

PURPOSE: To constrain the risk of severe toxicity in radiotherapy and radiosurgery, precise volume delineation of organs at risk is required. This task is still manually performed, which is time-consuming and prone to observer variability. To address these issues, and as alternative to atlas-based segmentation methods, machine learning techniques, such as support vector machines (SVM), have been recently presented to segment subcortical structures on magnetic resonance images (MRI). METHODS: SVM is proposed to segment the brainstem on MRI in multicenter brain cancer context. A dataset composed by 14 adult brain MRI scans is used to evaluate its performance. In addition to spatial and probabilistic information, five different image intensity values (IIVs) configurations are evaluated as features to train the SVM classifier. Segmentation accuracy is evaluated by computing the Dice similarity coefficient (DSC), absolute volumes difference (AVD) and percentage volume difference between automatic and manual contours. RESULTS: Mean DSC for all proposed IIVs configurations ranged from 0.89 to 0.90. Mean AVD values were below 1.5 cm(3), where the value for best performing IIVs configuration was 0.85 cm(3), representing an absolute mean difference of 3.99% with respect to the manual segmented volumes. CONCLUSION: Results suggest consistent volume estimation and high spatial similarity with respect to expert delineations. The proposed approach outperformed presented methods to segment the brainstem, not only in volume similarity metrics, but also in segmentation time. Preliminary results showed that the approach might be promising for adoption in clinical use.

Hepatic vessel segmentation for 3D planning of liver surgery experimental evaluation of a new fully automatic algorithm.

RATIONALE AND OBJECTIVES: The aim of this study was to identify the optimal parameter configuration of a new algorithm for fully automatic segmentation of hepatic vessels, evaluating its accuracy in view of its use in a computer system for three-dimensional (3D) planning of liver surgery. MATERIALS AND METHODS: A phantom reproduction of a human liver with vessels up to the fourth subsegment order, corresponding to a minimum diameter of 0.2 mm, was realized through stereolithography, exploiting a 3D model derived from a real human computed tomographic data set. Algorithm parameter configuration was experimentally optimized, and the maximum achievable segmentation accuracy was quantified for both single two-dimensional slices and 3D reconstruction of the vessel network, through an analytic comparison of the automatic segmentation performed on contrast-enhanced computed tomographic phantom images with actual model features. RESULTS: The optimal algorithm configuration resulted in a vessel detection sensitivity of 100% for vessels > 1 mm in diameter, 50% in the range 0.5 to 1 mm, and 14% in the range 0.2 to 0.5 mm. An average area overlap of 94.9% was obtained between automatically and manually segmented vessel sections, with an average difference of 0.06 mm(2). The average values of corresponding false-positive and false-negative ratios were 7.7% and 2.3%, respectively. CONCLUSIONS: A robust and accurate algorithm for automatic extraction of the hepatic vessel tree from contrast-enhanced computed tomographic volume images was proposed and experimentally assessed on a liver model, showing unprecedented sensitivity in vessel delineation. This automatic segmentation algorithm is promising for supporting liver surgery planning and for guiding intraoperative resections.

A new fully automatic and robust algorithm for fast segmentation of liver tissue and tumors from CT scans.

Accurate knowledge of the liver structure, including liver surface and lesion localization, is usually required in treatments such as liver tumor ablations and/or radiotherapy. This paper presents a new method and corresponding algorithm for fast segmentation of the liver and its internal lesions from CT scans. No interaction between the user and analysis system is required for initialization since the algorithm is fully automatic. A statistical model-based approach was created to distinguish hepatic tissue from other abdominal organs. It was combined to an active contour technique using gradient vector flow in order to obtain a smoother and more natural liver surface segmentation. Thereafter, automatic classification was performed to isolate hepatic lesions from liver parenchyma. Twenty-one datasets, presenting different anatomical and pathological situations, have been processed and analyzed. Special focus has been driven to the resulting processing time together with quality assessment. Our method allowed robust and efficient liver and lesion segmentations very close to the ground truth, in a relatively short processing time (average of 11.4 s for a 512 x 512-pixel slice). A volume overlap of 94.2% and an accuracy of 3.7 mm were achieved for liver surface segmentation. Sensitivity and specificity for tumor lesion detection were 82.6% and 87.5%, respectively.

Fully automatic liver segmentation through graph-cut technique.

The accurate knowledge of the liver structure including blood vessels topography, liver surface and lesion localizations is usually required in treatments like liver ablations and radiotherapy. In this paper, we propose an approach for automatic segmentation of liver complex geometries. It consists of applying a graph-cut method initialized by an adaptive threshold. The algorithm has been tested on 10 datasets (CT and MR). A parametric comparison with the results obtained by previous algorithms based on active contour is also carried out and discussed. Main limitations of active contour approaches result to be overcome and segmentation is improved. Feasibility to routinely use graph-cut approach for automatic liver segmentation is also demonstrated.

Automatic vertebral morphometry assessment.

Diagnosis of vertebral deformation and fracture is a key issue in the management of osteoporosis. Vertebral morphometry is the key solution to assess these vertebral deformities, but it is a tedious operation for the rheumatologist. In this paper, we propose an automatic approach to localize vertebrae and determine the vertebral morphometry. Based on a local phase symmetry measure, our approach shows promising results even on poor quality images. This work was tested on 22 scans of conventional lateral X-rays radiographs and compared to the results obtained by a trained radiologist. It resulted to have acceptable low error rates for localization and vertebral morphometry measures, but with a lack of robustness among images. Amelioration is necessary for images with an unoptimized patient positioning, when the X-ray beam is not perpendicular to the spine.