Weakly supervised object detection with 2D and 3D regression neural networks

Finding automatically multiple lesions in large images is a common problem in medical image analysis. Solving this problem can be challenging if, during optimization, the automated method cannot access information about the location of the lesions nor is given single examples of the lesions. We propose a new weakly supervised detection method using neural networks, that computes attention maps revealing the locations of brain lesions. These attention maps are computed using the last feature maps of a segmentation network optimized only with global image-level labels. The proposed method can generate attention maps at full input resolution without need for interpolation during preprocessing, which allows small lesions to appear in attention maps. For comparison, we modify state-of-the-art methods to compute attention maps for weakly supervised object detection, by using a global regression objective instead of the more conventional classification objective. This regression objective optimizes the number of occurrences of the target object in an image, e.g. the number of brain lesions in a scan, or the number of digits in an image. We study the behavior of the proposed method in MNIST-based detection datasets, and evaluate it for the challenging detection of enlarged perivascular spaces – a type of brain lesion – in a dataset of 2202 3D scans with point-wise annotations in the center of all lesions in four brain regions. In MNIST-based datasets, the proposed method outperforms the other methods. In the brain dataset, the weakly supervised detection methods come close to the human intrarater agreement in each region. The proposed method reaches the best area under the curve in two out of four regions, and has the lowest number of false positive detections in all regions, while its average sensitivity over all regions is similar to that of the other best methods. The proposed method can facilitate epidemiological and clinical studies of enlarged perivascular spaces and help advance research in the etiology of enlarged perivascular spaces and in their relationship with cerebrovascular diseases.     

IDR‐1018 induces cell proliferation,migration, and reparative gene expression in 2D culture and 3D human skin equivalents

Skin lesions are associated with functional/cosmetic problems for those afflicted. Scarless regeneration is a challenge, not limited to the skin, and focus of active investigation. Recently, the host defense peptide innate defense regulatory peptide 1018 (IDR‐1018) has shown exciting regenerative properties. Nevertheless, literature regarding IDR‐1018 regenerative potential is scarce and limited to animal models. Here, we evaluated the regenerative potential of IDR‐1018 using human 2D and 3D human skin equivalents. First, we investigated IDR‐1018 using human cells found in skin—primary fibroblasts, primary keratinocytes, and the MeWo melanocytes cell line. IDR‐1018 promoted cell proliferation and expression of marker of proliferation Ki‐67, matrix metalloproteinase 1, and hyaluronan synthase 2 by fibroblasts. In keratinocytes, a drastic increase in expression was observed for Ki‐67, matrix metalloproteinase 1, C‐X‐C motif chemokine receptor type 4, C‐X‐C motif chemokine receptor type 7, fibroblast growth factor 2, hyaluronan synthase 2, vascular endothelial growth factor, and elastin, reflecting an intense stimulation of these cells. In melanocytes, increased migration and proliferation were observed following IDR‐1018 treatment. The capacity of IDR‐1018 to promote dermal contraction was verified using a dermal model. Finally, using a 3D human skin equivalent lesion model, we revealed that the regenerative potential of IDR1018, previously tested in mice and pigs, is valid for human skin tissue. Lesions closed faster in IDR‐1018‐treated samples, and the gene expression signature observed in 2D was reproduced in the 3D human skin equivalents. Overall, the present data show the regenerative potential of IDR‐1018 in an experimental system comprising human cells, underscoring the potential application for clinical investigation.     

Aging decreases the capacity of human skin to produce vitamin D3.

An evaluation of surgically obtained skin (age range, 8-92 yr) revealed that there is an age-dependent decrease in the epidermal concentrations of provitamin D3 (7-dehydrocholesterol). To ascertain that aging indeed decreased the capacity of human skin to produce vitamin D3, some of the skin samples were exposed to ultraviolet radiation and the content of previtamin D3 was determined in the epidermis and dermis. The epidermis in the young and older subjects was the major site for the formation of previtamin D3, accounting for greater than 80% of the total previtamin D3 that was produced in the skin. A comparison of the amount of previtamin D3 produced in the skin from the 8- and 18-yr-old subjects with the amount produced in the skin from the 77- and 82-yr-old subjects revealed that aging can decrease by greater than twofold the capacity of the skin to produce previtamin D3. Recognition of this difference may be extremely important for the elderly, who infrequently expose a small area of skin to sunlight and who depend on this exposure for their vitamin D nutritional needs.     

Liver lesion changes analysis in longitudinal CECT scans by simultaneous deep learning voxel classification with SimU-Net

The identification and quantification of liver lesions changes in longitudinal contrast enhanced CT (CECT) scans is required to evaluate disease status and to determine treatment efficacy in support of clinical decision-making. This paper describes a fully automatic end-to-end pipeline for liver lesion changes analysis in consecutive (prior and current) abdominal CECT scans of oncology patients. The three key novelties are: (1) SimU-Net, a simultaneous multi-channel 3D R2U-Net model trained on pairs of registered scans of each patient that identifies the liver lesions and their changes based on the lesion and healthy tissue appearance differences; (2) a model-based bipartite graph lesions matching method for the analysis of lesion changes at the lesion level; (3) a method for longitudinal analysis of one or more of consecutive scans of a patient based on SimU-Net that handles major liver deformations and incorporates lesion segmentations from previous analysis. To validate our methods, five experimental studies were conducted on a unique dataset of 3491 liver lesions in 735 pairs from 218 clinical abdominal CECT scans of 71 patients with metastatic disease manually delineated by an expert radiologist. The pipeline with the SimU-Net model, trained and validated on 385 pairs and tested on 249 pairs, yields a mean lesion detection recall of 0.86±0.14, a precision of 0.74±0.23 and a lesion segmentation Dice of 0.82±0.14 for lesions > 5 mm. This outperforms a reference standalone 3D R2-UNet mdel that analyzes each scan individually by ∼50% in precision with similar recall and Dice score on the same training and test datasets. For lesions matching, the precision is 0.86±0.18 and the recall is 0.90±0.15. For lesion classification, the specificity is 0.97±0.07, the precision is 0.85±0.31, and the recall is 0.86±0.23. Our new methods provide accurate and comprehensive results that may help reduce radiologists' time and effort and improve radiological oncology evaluation.     

基于MRI图像的主动脉分割与三维建模

目的基于MRI图像序列建立主动脉的三维几何模型并进行计算网格的划分,以用于主动脉血流动力学特性的模拟。方法采用心电R波触发和呼吸控制的方式在体扫描得到心动周期20个时相760幅MRI图像,利用Mimics软件对所获取的图像序列进行图像预处理、分割和三维重建,然后将所建立的三维模型导入到ADINA软件中进行计算网格的划分。结果建立了20个主动脉三维模型,分别代表主动脉在心动周期不同时相的状态,同时,还实现了计算网格的划分。结论该方法可得到进行主动脉血流动力学仿真所需的三维几何模型和计算网格;同时,该方法也可用于人体其他组织的三维建模和网格划分。     

Evaluating thin compression paddles for mammographically compatible ultrasound

We are developing a combined digital mammography/3D ultrasound system to improve detection and/or characterization of breast lesions. Ultrasound scanning through a mammographic paddle could significantly reduce signal level, degrade beam focusing and create reverberations. Thus, appropriate paddle choice is essential for accurate sonographic lesion detection and assessment with this system. In this study, we characterized ultrasound image quality through paddles of varying materials (lexan, polyurethane, TPX, mylar) and thicknesses (0.25 to 2.5 mm). Analytical experiments focused on lexan and TPX, which preliminary results demonstrated were most competitive. Spatial and contrast resolution, side-lobe and range lobe levels, contrast and signal strength were compared with no-paddle images. When the beamforming of the system was corrected to account for imaging through the paddle, the TPX 2.5 mm paddle performed the best. Test objects imaged through this paddle demonstrated < or = 15% reduction in spatial resolution, < or = 7.5 dB signal loss, < or = 3 dB contrast loss and range lobe levels > or = 35 dB below signal maximum over 4 cm. TPX paddles <2.5 mm could also be used with this system, depending on imaging goals. In 10 human subjects with cysts, small CNR losses were observed but were determined to be statistically insignificant. Radiologists concluded that 75% of cysts in through-paddle scans were at least as detectable as in their corresponding direct-contact scans.     

CAN3D: Fast 3D medical image segmentation via compact context aggregation

Direct automatic segmentation of objects in 3D medical imaging, such as magnetic resonance (MR) imaging, is challenging as it often involves accurately identifying multiple individual structures with complex geometries within a large volume under investigation. Most deep learning approaches address these challenges by enhancing their learning capability through a substantial increase in trainable parameters within their models. An increased model complexity will incur high computational costs and large memory requirements unsuitable for real-time implementation on standard clinical workstations, as clinical imaging systems typically have low-end computer hardware with limited memory and CPU resources only. This paper presents a compact convolutional neural network (CAN3D) designed specifically for clinical workstations and allows the segmentation of large 3D Magnetic Resonance (MR) images in real-time. The proposed CAN3D has a shallow memory footprint to reduce the number of model parameters and computer memory required for state-of-the-art performance and maintain data integrity by directly processing large full-size 3D image input volumes with no patches required. The proposed architecture significantly reduces computational costs, especially for inference using the CPU. We also develop a novel loss function with extra shape constraints to improve segmentation accuracy for imbalanced classes in 3D MR images. Compared to state-of-the-art approaches (U-Net3D, improved U-Net3D and V-Net), CAN3D reduced the number of parameters up to two orders of magnitude and achieved much faster inference, up to 5 times when predicting with a standard commercial CPU (instead of GPU). For the open-access OAI-ZIB knee MR dataset, in comparison with manual segmentation, CAN3D achieved Dice coefficient values of (mean = 0.87 ± 0.02 and 0.85 ± 0.04) with mean surface distance errors (mean = 0.36 ± 0.32 mm and 0.29 ± 0.10 mm) for imbalanced classes such as (femoral and tibial) cartilage volumes respectively when training volume-wise under only 12G video memory. Similarly, CAN3D demonstrated high accuracy and efficiency on a pelvis 3D MR imaging dataset for prostate cancer consisting of 211 examinations with expert manual semantic labels (bladder, body, bone, rectum, prostate) now released publicly for scientific use as part of this work.     

Current status of 3D/4D volume ultrasound of the breast

3D/4D volume ultrasound is an established method that offers various options for analyzing and presenting ultrasound volume data. The following imaging techniques are based on automatically acquired ultrasound volumes. The multiplanar view is the typical mode of 3D ultrasound data presentation. The niche mode view is a cut open view of the volume data set. The surface mode is a rendering technique that represents the data within a volume of interest (VOI) with different slice thicknesses (typically 1-4 mm) with a contrast-enhanced surface algorithm. Related to the diagnostic target, the transparency mode helps to present echopoor or echorich structures and their spatial relationships within the ultrasound volume. Glass body rendering is a special type of transparency mode that makes the grayscale data transparent and shows the color flow data in a surface render mode. The inversion mode offers a three-dimensional surface presentation of echopoor lesions. Volume Contrast Imaging (VCI) works with static 3D volume data and is able to be used with 4D for dynamic scanning. Volume calculation of a lesion and virtual computer-assisted organ analysis of the same lesion is performed with VoCal software. Tomographic Ultrasound Imaging (TUI) is the perfect tool to document static 3D ultrasound volumes. 3D/4D volume ultrasound of the breast provides diagnostic information of the coronal plane. In this plane benign lesions show the compression pattern sign, while malignant lesions show the retraction pattern or star pattern sign. The indeterminate pattern of a lesion combines signs of compression and retraction or star pattern in the coronal plane. Glass body rendering in combination with Power-Doppler, Color-Doppler or High-Definition Flow Imaging presents the intra- and peritumoral three-dimensional vascular architecture. 3D targeting shows correct or incorrect needle placement in all three planes after 2D or 4D needle guidance. In conclusion, it is safe to say that 3D/4D volume ultrasound of the breast is technically advanced and suitable for daily diagnostic and interventional breast work in addition to routinely used 2D sonography.     

From colour to tissue histology: Physics-based interpretation of images of pigmented skin lesions

Through an understanding of the image formation process, diagnostically important facts about the internal structure and composition of pigmented skin lesions can be derived from their colour images. A physics-based model of tissue colouration provides a cross-reference between image colours and the underlying histological parameters. It is constructed by computing the spectral composition of light remitted from the skin given parameters specifying its structure and optical properties. The model is representative of all the normal human skin colours, irrespective of racial origin, age or gender. Abnormal skin colours do not conform to this model and thus can be detected. Once the model is constructed, for each pixel in a colour image its histological parameters are computed from the model. Represented as images, these 'parametric maps' show the concentration of dermal and epidermal melanin, blood and collagen thickness across the imaged skin as well as locations where abnormal colouration exists. In a clinical study the parametric maps were used by a clinician to detect the presence of malignant melanoma in a set of 348 pigmented lesions imaged using a commercial device, the SIAscope. Logistic regression identified the presence of melanin in the dermis, the abnormal distribution of blood within the lesion and the lesion size as the most diagnostically informative features. Classification based on these features showed 80.1% sensitivity and 82.7% specificity in melanoma detection.     

3．0T MR肝脏三维容积超快速多期动态增强扫描在肝局灶性病变诊断中的价值

目的：探讨肝脏三维容积超快速多期动态增强扫描（3D CE LAVA）对肝占位性病变的诊断价值。资料与方法：对76例经病理证实的肝脏肿瘤患者行三维容积超快速多期动态增强扫描和2D GRE T1WI增强扫描，并采用最大强度投影（MIP）和MPR方式进行图像重建，统计病变显示率及其强化程度。结果：全肝动脉期多时相三维动态增强MR扫描及其图像重建，可同时显示肝脏肿瘤的动态强化过程和肝血管形态，对肝脏肿瘤的诊断及鉴别诊断提供依据。结论：MRI 3D LAVA多期动态增强扫描无论在病灶的显示及定性诊断方面，还是在病灶血管及肝血管解剖的显示上均较2D GRE T1WI增强扫描具有较高的临床价值。     

3D regression neural network for the quantification of enlarged perivascular spaces in brain MRI

Enlarged perivascular spaces (EPVS) in the brain are an emerging imaging marker for cerebral small vessel disease, and have been shown to be related to increased risk of various neurological diseases, including stroke and dementia. Automated quantification of EPVS would greatly help to advance research into its etiology and its potential as a risk indicator of disease. We propose a convolutional network regression method to quantify the extent of EPVS in the basal ganglia from 3D brain MRI. We first segment the basal ganglia and subsequently apply a 3D convolutional regression network designed for small object detection within this region of interest. The network takes an image as input, and outputs a quantification score of EPVS. The network has significantly more convolution operations than pooling ones and no final activation, allowing it to span the space of real numbers. We validated our approach using a dataset of 2000 brain MRI scans scored visually. Experiments with varying sizes of training and test sets showed that a good performance can be achieved with a training set of only 200 scans. With a training set of 1000 scans, the intraclass correlation coefficient (ICC) between our scoring method and the expert’s visual score was 0.74. Our method outperforms by a large margin - more than 0.10 - four more conventional automated approaches based on intensities, scale-invariant feature transform, and random forest. We show that the network learns the structures of interest and investigate the influence of hyper-parameters on the performance. We also evaluate the reproducibility of our network using a set of 60 subjects scanned twice (scan-rescan reproducibility). On this set our network achieves an ICC of 0.93, while the intrarater agreement reaches 0.80. Furthermore, the automated EPVS scoring correlates similarly to age as visual scoring.     

Probabilistic 3D MRI atlas of the human cerebellar dentate/interposed nuclei

In a previous study, a three-dimensional (3D) MRI atlas of the human cerebellar nuclei was introduced based on findings in one healthy human subject [Dimitrova, A., Weber, J., Redies, C., Kindsvater, K., Maschke, M., Kolb, F.P., Forsting, M., Diener, H.C., Timmann, D., 2002. MRI atlas of the human cerebellar nuclei. NeuroImage 17, 240-255]. The present MRI investigation was designed to study variability of the anatomy of the dentate/interposed nuclei in a larger group of healthy subjects. Similar to our previous study, iron-induced susceptibility artifacts were used to visualize the cerebellar nuclei as hypointensities on MR images. Data of 63 healthy subjects (27 female, 36 male; mean age 45.3+/-13.4 years, age range 22--71 years) were included. A 3D axial volume of the cerebellum was acquired using a T2*-weighted FLASH sequence on a Siemens Sonata 1.5 T MR scanner. Each volume was registered, re-sampled to 1.00 x 1.00 x 1.00 mm(3) voxel size and spatially normalized into a standard proportional stereotaxic space using SPM99. Dentate/interposed nuclei were traced on axial images and saved as regions of interest using MRIcro-software by two independent examiners. A probabilistic 3D MRI atlas of the cerebellar dentate/interposed nuclei is presented based on findings in all subjects.     

实时三维超声心动图在心脏瓣膜病中的应用

目的探讨实时三维超声心动图在心脏瓣膜病诊断中的作用,比较其与二维超声心动图的不同功能。方法总结了本院2003年6月至2004年9月间接受实时三维超声心动图检查并经手术证实为心脏瓣膜病患者30例的超声检查情况。结果实时三维超声心动图所能提示的情况较二维超声心动图提示的内容为多,且可较明确地定位病变的位置。结论实时三维超声心动图在心脏瓣膜病的诊断中可对二维超声心动图做出有益的补充,对外科瓣膜手术方案的制定具有指导意义。实时三维超声心动图具有比较广阔的临床应用前景。     

Efficient and robust model-to-image alignment using 3D scale-invariant features

This paper presents feature-based alignment (FBA), a general method for efficient and robust model-to-image alignment. Volumetric images, e.g. CT scans of the human body, are modeled probabilistically as a collage of 3D scale-invariant image features within a normalized reference space. Features are incorporated as a latent random variable and marginalized out in computing a maximum a posteriori alignment solution. The model is learned from features extracted in pre-aligned training images, then fit to features extracted from a new image to identify a globally optimal locally linear alignment solution. Novel techniques are presented for determining local feature orientation and efficiently encoding feature intensity in 3D. Experiments involving difficult magnetic resonance (MR) images of the human brain demonstrate FBA achieves alignment accuracy similar to widely-used registration methods, while requiring a fraction of the memory and computation resources and offering a more robust, globally optimal solution. Experiments on CT human body scans demonstrate FBA as an effective system for automatic human body alignment where other alignment methods break down.     

Learning to map 2D ultrasound images into 3D space with minimal human annotation

In fetal neurosonography, aligning two-dimensional (2D) ultrasound scans to their corresponding plane in the three-dimensional (3D) space remains a challenging task. In this paper, we propose a convolutional neural network that predicts the position of 2D ultrasound fetal brain scans in 3D atlas space. Instead of purely supervised learning that requires heavy annotations for each 2D scan, we train the model by sampling 2D slices from 3D fetal brain volumes, and target the model to predict the inverse of the sampling process, resembling the idea of self-supervised learning.We propose a model that takes a set of images as input, and learns to compare them in pairs. The pairwise comparison is weighted by the attention module based on its contribution to the prediction, which is learnt implicitly during training. The feature representation for each image is thus computed by incorporating the relative position information to all the other images in the set, and is later used for the final prediction.We benchmark our model on 2D slices sampled from 3D fetal brain volumes at 18–22 weeks' gestational age. Using three evaluation metrics, namely, Euclidean distance, plane angles and normalized cross correlation, which account for both the geometric and appearance discrepancy between the ground-truth and prediction, in all these metrics, our model outperforms a baseline model by as much as 23%, when the number of input images increases. We further demonstrate that our model generalizes to (i) real 2D standard transthalamic plane images, achieving comparable performance as human annotations, as well as (ii) video sequences of 2D freehand fetal brain scans.     

Assessing cerebrospinal fluid flow connectivity using 3D gradient echo phase contrast velocity encoded MRI

The aim of this study is to investigate the feasibility of using three-directional velocity encoded 3D gradient echo (GE) phase contrast (PC) imaging to assess cerebrospinal fluid (CSF) flow connectivity in the human brain. Five healthy volunteers were scanned using low velocity sensitivity (V(enc) = 0.04-0.05 m s(-1)). Flow-time curves were compared to standard 2D PC scans. The 3D data were used to reconstruct in vivo CSF flow volumes based on time-averaged phase-difference information, and the patency of the CSF flow pathways was assessed using nearest-neighbour connectivity. A pulsatile flow phantom was used to gauge the measurement accuracy of the CSF flow volumes at low flow velocities. Flow connectivity from the lateral ventricles down to the cisterna magna was successfully demonstrated in all volunteers. The phantom tests showed a good distinction between the flow cavities and the background noise. 3D PC imaging results in CSF flow waveforms with similar pulsatility but underestimated peak velocities compared to 2D PC data. 3D time-resolved velocity encoded GE imaging has successfully been applied to assess CSF flow connectivity in normal subjects.     

RecON: Online learning for sensorless freehand 3D ultrasound reconstruction

Sensorless freehand 3D ultrasound (US) reconstruction based on deep networks shows promising advantages, such as large field of view, relatively high resolution, low cost, and ease of use. However, existing methods mainly consider vanilla scan strategies with limited inter-frame variations. These methods thus are degraded on complex but routine scan sequences in clinics. In this context, we propose a novel online learning framework for freehand 3D US reconstruction under complex scan strategies with diverse scanning velocities and poses. First, we devise a motion-weighted training loss in training phase to regularize the scan variation frame-by-frame and better mitigate the negative effects of uneven inter-frame velocity. Second, we effectively drive online learning with local-to-global pseudo supervisions. It mines both the frame-level contextual consistency and the path-level similarity constraint to improve the inter-frame transformation estimation. We explore a global adversarial shape before transferring the latent anatomical prior as supervision. Third, we build a feasible differentiable reconstruction approximation to enable the end-to-end optimization of our online learning. Experimental results illustrate that our freehand 3D US reconstruction framework outperformed current methods on two large, simulated datasets and one real dataset. In addition, we applied the proposed framework to clinical scan videos to further validate its effectiveness and generalizability.     

Exudate-based diabetic macular edema detection in fundus images using publicly available datasets

Diabetic macular edema (DME) is a common vision threatening complication of diabetic retinopathy. In a large scale screening environment DME can be assessed by detecting exudates (a type of bright lesions) in fundus images. In this work, we introduce a new methodology for diagnosis of DME using a novel set of features based on colour, wavelet decomposition and automatic lesion segmentation. These features are employed to train a classifier able to automatically diagnose DME through the presence of exudation. We present a new publicly available dataset with ground-truth data containing 169 patients from various ethnic groups and levels of DME. This and other two publicly available datasets are employed to evaluate our algorithm. We are able to achieve diagnosis performance comparable to retina experts on the MESSIDOR (an independently labelled dataset with 1200 images) with cross-dataset testing (e.g., the classifier was trained on an independent dataset and tested on MESSIDOR). Our algorithm obtained an AUC between 0.88 and 0.94 depending on the dataset/features used. Additionally, it does not need ground truth at lesion level to reject false positives and is computationally efficient, as it generates a diagnosis on an average of 4.4 s (9.3 s, considering the optic nerve localisation) per image on an 2.6 GHz platform with an unoptimised Matlab implementation.     

A CAD system for the 3D location of lesions in mammograms

A CAD system for estimating the 3D (three-dimensional) positions of lesions found in two mammographic views is described. The system is an extension of our previous method [Comput. Vis. Image Understand. 83 (2001) 38] which finds corresponding 2D positions in different mammographic views. The method calculates curved epipolar lines by developing a simulation of breast deformation into stereo camera geometry. Using such curved epipolar lines, not only can we determine point correspondences, but can estimate the 3D location of a lesion. In this paper, we first explain the underlying principles and system organisation. The correctness of the 3D positions calculated by the system is examined using a set of breast lesions, which appear both in mammograms and in MRI data. The experimental results demonstrate the clinical promise of the CAD system.     

A fully automatic end-to-end method for content-based image retrieval of CT scans with similar liver lesion annotations

Purpose

The goal of medical content-based image retrieval (M-CBIR) is to assist radiologists in the decision-making process by retrieving medical cases similar to a given image. One of the key interests of radiologists is lesions and their annotations, since the patient treatment depends on the lesion diagnosis. Therefore, a key feature of M-CBIR systems is the retrieval of scans with the most similar lesion annotations. To be of value, M-CBIR systems should be fully automatic to handle large case databases.

Methods

We present a fully automatic end-to-end method for the retrieval of CT scans with similar liver lesion annotations. The input is a database of abdominal CT scans labeled with liver lesions, a query CT scan, and optionally one radiologist-specified lesion annotation of interest. The output is an ordered list of the database CT scans with the most similar liver lesion annotations. The method starts by automatically segmenting the liver in the scan. It then extracts a histogram-based features vector from the segmented region, learns the features’ relative importance, and ranks the database scans according to the relative importance measure. The main advantages of our method are that it fully automates the end-to-end querying process, that it uses simple and efficient techniques that are scalable to large datasets, and that it produces quality retrieval results using an unannotated CT scan.

Results

Our experimental results on 9 CT queries on a dataset of 41 volumetric CT scans from the 2014 Image CLEF Liver Annotation Task yield an average retrieval accuracy (Normalized Discounted Cumulative Gain index) of 0.77 and 0.84 without/with annotation, respectively.

Conclusions

Fully automatic end-to-end retrieval of similar cases based on image information alone, rather that on disease diagnosis, may help radiologists to better diagnose liver lesions.