Efficiently compressing 3D medical images for teleinterventions via CNNs and anisotropic diffusion.

PURPOSE: Efficient compression of images while preserving image quality has the potential to be a major enabler of effective remote clinical diagnosis and treatment, since poor Internet connection conditions are often the primary constraint in such services. This paper presents a framework for organ-specific image compression for teleinterventions based on a deep learning approach and anisotropic diffusion filter. METHODS: The proposed method, deep learning and anisotropic diffusion (DLAD), uses a convolutional neural network architecture to extract a probability map for the organ of interest; this probability map guides an anisotropic diffusion filter that smooths the image except at the location of the organ of interest. Subsequently, a compression method, such as BZ2 and HEVC-visually lossless, is applied to compress the image. We demonstrate the proposed method on three-dimensional (3D) CT images acquired for radio frequency ablation (RFA) of liver lesions. We quantitatively evaluate the proposed method on 151 CT images using peak-signal-to-noise ratio ( PSNR ), structural similarity ( SSIM ), and compression ratio ( CR ) metrics. Finally, we compare the assessments of two radiologists on the liver lesion detection and the liver lesion center annotation using 33 sets of the original images and the compressed images. RESULTS: The results show that the method can significantly improve CR of most well-known compression methods. DLAD combined with HEVC-visually lossless achieves the highest average CR of 6.45, which is 36% higher than that of the original HEVC and outperforms other state-of-the-art lossless medical image compression methods. The means of PSNR and SSIM are 70 dB and 0.95, respectively. In addition, the compression effects do not statistically significantly affect the assessments of the radiologists on the liver lesion detection and the lesion center annotation. CONCLUSIONS: We thus conclude that the method has a high potential to be applied in teleintervention applications.

Quantification of nonrigid liver deformation in radiofrequency ablation interventions using image registration.

Multimodal image fusion for image guidance in minimally invasive liver interventions generally requires the registration of pre-operatively acquired images with interventional images of the patient. Whereas rigid registration approaches are fast and can be used in an interventional setting, the actual liver deformation may be nonrigid. The purpose of this paper is to assess the magnitude of nonrigid deformation of the liver between pre-operative and interventional CT images in the case of tumor ablations, over the full liver and over parts of the liver that match the volumes typically imaged by a 3D ultrasound transducer. We acquired 3D abdominal CT scans of 38 patients that had undergone the radiofrequency ablation of liver tumors, pre-operative CT images as well as intraoperative CT images. To determine the magnitude of liver deformation due to pose changes and respiration, we nonrigidly registered the pre-operative CT scan with the intraoperative CT scan. By fitting the deformation to a rigid transformation in the region of interest and computing the residual displacements, the nonrigid deformation part can be quantified. We performed quantifications over the complete liver, as well as for two volumes of interest representative of sub-xiphoidal and intercostal 3D ultrasound acquisitions. The results showed that a substantial amount of nonrigid deformation was found, and rotation of the patient's pose and deep inhalation caused significant liver deformation. Hence we concluded that nonrigid motion correction in the interventions should be taken into account.

Accuracy of semi-automated versus manual localisation of liver tumours in CT-guided ablation procedures.

OBJECTIVES: To compare the accuracy of liver tumour localisation in intraprocedural computed tomography (CT) images of computer-based rigid registration or non-rigid registration versus mental registration performed by interventional radiologists. METHODS: Retrospectively (2009-2017), 35 contrast-enhanced CT (CECT) images incorporating 56 tumours, acquired during CT-guided ablation procedures and their corresponding pre-procedural diagnostic CECTs were retrieved from the picture archiving and communication system (PACS). The original intraprocedural CECTs were de-enhanced to create a virtually unenhanced CT image (VUCT). Alignment of diagnostic CECTs to their corresponding intraprocedural VUCTs was performed with non-rigid or rigid registration. Mental registration was performed by four interventional radiologists. The original intraprocedural CECT served as the reference standard. Accuracy of tumour localisation was assessed with the target registration error (TRE). Statistical differences were analysed with the Wilcoxon signed-rank test. RESULTS: Non-rigid registration failed to register two CT datasets, incorporating four tumours. In the remaining 33 datasets, non-rigid, rigid and mental registration showed a median TRE of 3.9 mm, 9.0 mm and 10.9 mm, respectively. Non-rigid registration was significantly more accurate in tumour centre localisation in comparison to rigid (p < 0.001) or mental registration (p < 0.001). Rigid registration was not statistically different from mental registration (p = 0.169). Non-rigid registration was most accurate in localising tumour centres in 42 out of 52 tumours (80.8%), while rigid and mental registration were most accurate in only seven (13.5%) and three (5.8%) tumours, respectively. CONCLUSIONS: Computer-based non-rigid registration is statistically significantly more accurate in localising liver tumours in intraprocedural unenhanced CT images in comparison to rigid registration or interventional radiologists' mental mapping abilities. KEY POINTS: • Computer-based non-rigid registration is better (p < 0.001) in localising target tumours prior to ablation in intraprocedural CT images in comparison to rigid registration or interventional radiologists' mental mapping abilities. • Human experts perform sub-optimal localisation of target tumours when relying solely on mental mapping during challenging CT-guided procedures. • This non-rigid registration method shows promising results as a safe alternative to intravenous contrast media in liver tumour localisation prior to ablation during CT-guided procedures.

Non-Rigid Registration of Liver CT Images for CT-Guided Ablation of Liver Tumors.

CT-guided percutaneous ablation for liver cancer treatment is a relevant technique for patients not eligible for surgery and with tumors that are inconspicuous on US imaging. The lack of real-time imaging and the use of a limited amount of CT contrast agent make targeting the tumor with the needle challenging. In this study, we evaluate a registration framework that allows the integration of diagnostic pre-operative contrast enhanced CT images and intra-operative non-contrast enhanced CT images to improve image guidance in the intervention. The liver and tumor are segmented in the pre-operative contrast enhanced CT images. Next, the contrast enhanced image is registered to the intra-operative CT images in a two-stage approach. First, the contrast-enhanced diagnostic image is non-rigidly registered to a non-contrast enhanced image that is conventionally acquired at the start of the intervention. In case the initial registration is not sufficiently accurate, a refinement step is applied using non-rigid registration method with a local rigidity term. In the second stage, the intra-operative CT-images that are used to check the needle position, which often consist of only a few slices, are registered rigidly to the intra-operative image that was acquired at the start of the intervention. Subsequently, the diagnostic image is registered to the current intra-operative image, using both transformations, this allows the visualization of the tumor region extracted from pre-operative data in the intra-operative CT images containing needle. The method is evaluated on imaging data of 19 patients at the Erasmus MC. Quantitative evaluation is performed using the Dice metric, mean surface distance of the liver border and corresponding landmarks in the diagnostic and the intra-operative images. The registration of the diagnostic CT image to the initial intra-operative CT image did not require a refinement step in 13 cases. For those cases, the resulting registration had a Dice coefficient for the livers of 91.4%, a mean surface distance of 4.4 mm and a mean distance between corresponding landmarks of 4.7 mm. For the three cases with a refinement step, the registration result significantly improved (p<0.05) compared to the result of the initial non rigid registration method (DICE of 90.3% vs 71.3% and mean surface distance of 5.1 mm vs 11.3 mm and mean distance between corresponding landmark of 6.4 mm vs 10.2 mm). The registration of the preoperative data with the needle image in 16 cases yielded a DICE of 90.1% and a mean surface distance of 5.2 mm. The remaining three cases with DICE smaller than 80% were classified as unsuccessful registration. The results show that this is promising tool for liver image registration in interventional radiology.

An automatic registration method for pre- and post-interventional CT images for assessing treatment success in liver RFA treatment.

PURPOSE: In image-guided radio frequency ablation for liver cancer treatment, pre- and post-interventional CT images are typically used to verify the treatment success of the therapy. In current clinical practice, the tumor zone in the diagnostic, preinterventional images is mentally or manually mapped to the ablation zone in the post-interventional images to decide success of the treatment. However, liver deformation and differences in image quality as well as in texture of the ablation zone and the tumor area make the mental or manual registration a challenging task. Purpose of this paper is to develop an automatic framework to register the pre-interventional image to the post-interventional image. METHODS: The authors propose a registration approach enabling a nonrigid deformation of the tumor to the ablation zone, while keeping locally rigid deformation of the tumor area. The method was evaluated on CT images of 38 patient datasets from Erasmus MC. The evaluation is based on Dice coefficients of the liver segmentation on both the pre-interventional and post-interventional images, and mean distances between the liver segmentations. Additionally, residual distances after registration between corresponding landmarks and local mean surface distance in the images were computed. RESULTS: The results show that rigid registration gives a Dice coefficient of 87.9%, a mean distance of the liver surfaces of 5.53 mm, and a landmark error of 5.38 mm, while non-rigid registration with local rigid deformation has a Dice coefficient of 92.2%, a mean distance between the liver segmentation boundaries near the tumor area of 3.83 mm, and a landmark error of 2.91 mm, where a part of this error can be attributed to the slice spacing in the authors' CT images. CONCLUSIONS: This method is thus a promising tool to assess the success of RFA liver cancer treatment.

Quantitative evaluation of noise reduction and vesselness filters for liver vessel segmentation on abdominal CTA images.

Liver vessel segmentation in CTA images is a challenging task, especially in the case of noisy images. This paper investigates whether pre-filtering improves liver vessel segmentation in 3D CTA images. We introduce a quantitative evaluation of several well-known filters based on a proposed liver vessel segmentation method on CTA images. We compare the effect of different diffusion techniques i.e. Regularized Perona-Malik, Hybrid Diffusion with Continuous Switch and Vessel Enhancing Diffusion as well as the vesselness approaches proposed by Sato, Frangi and Erdt. Liver vessel segmentation of the pre-processed images is performed using a histogram-based region grown with local maxima as seed points. Quantitative measurements (sensitivity, specificity and accuracy) are determined based on manual landmarks inside and outside the vessels, followed by T-tests for statistic comparisons on 51 clinical CTA images. The evaluation demonstrates that all the filters make liver vessel segmentation have a significantly higher accuracy than without using a filter (p < 0.05); Hybrid Diffusion with Continuous Switch achieves the best performance. Compared to the diffusion filters, vesselness filters have a greater sensitivity but less specificity. In addition, the proposed liver vessel segmentation method with pre-filtering is shown to perform robustly on a clinical dataset having a low contrast-to-noise of up to 3 (dB). The results indicate that the pre-filtering step significantly improves liver vessel segmentation on 3D CTA images.