Thyroid disorders are a complex group of diseases that require an accurate diagnosis for effective treatment. Fine-needle aspiration biopsies can assist in detecting many thyroid diseases. These materials can be analyzed visually using traditional computer vision methods, despite the limitations of complex samples. To address this problem, we propose a novel approach that uses hyperspectral imaging (HSI) to analyze thyroid biological samples. HSI measures the absorbance of infrared light by biological samples using a micro Fourier transform infrared spectroscopy (micro-FTIR) and converts this data into hyperspectral images. In this study, we used HSI to train and validate a recurrent neural network to classify thyroid samples as healthy, cancerous, or goiter. Our experiments, based on the k-fold cross-validation, achieved an overall accuracy of 96.88%, a sensitivity of 96.87%, and a specificity of 98.45%. These results demonstrate the potential of hyperspectral imaging as a tool to assist pathologists in the diagnosis of thyroid disease.

Nowadays the measurements of chronic wounds are mostly performed manually by clinicians, despite on high limitations of such method. Modern technologies make it possible to do accurate wound assessment. However, due to high costs and limited availability of special equipment, such methods are not widely used. Here, we present a concept of a low-cost wound assessment system, which, with a help of computer vision techniques, will create dense 3D reconstructions of wounds from images taken with a hand-held camera. Our system will also provide wound segmentation and will allow wound tracking over time. Additional color correction will be applied to support qualitative analysis of wounds.

This work was

OBJECTIVE: Morbidity and mortality due to preeclampsia in settings with limited resources often results from delayed diagnosis. The Congo Red Dot (CRD) test, a simple modality to assess the presence of misfolded proteins in urine, shows promise as a diagnostic and prognostic tool for preeclampsia. We propose an innovative mobile health (mHealth) solution that enables the quantification of the CRD test as a batch laboratory test, with minimal cost and equipment. METHODS: A smartphone application that guides the user through seven easy steps, and that can be used successfully by non-specialized personnel, was developed. After image acquisition, a robust analysis runs on a smartphone, quantifying the CRD test response without the need for an internet connection or additional hardware. In the first stage, the basic image processing algorithms and supporting test standardizations were developed using urine samples from 218 patients. In the second stage, the standardized procedure was evaluated on 328 urine specimens from 273 women. In the third stage, the application was tested for robustness using four different operators and 94 altered samples. RESULTS: In the first stage, the image processing chain was set up with high correlation to manual analysis (z-test P < 0.001). In the second stage, a high agreement between manual and automated processing was calculated (Lin's concordance coefficient ρc = 0.968). In the last stage, sources of error were identified and remedies were developed accordingly. Altered samples resulted in an acceptable concordance with the manual gold-standard (Lin's ρc = 0.914). CONCLUSION: Combining smartphone-based image analysis with molecular-specific disease features represents a cost-effective application of mHealth that has the potential to fill gaps in access to health care solutions that are critical to reducing adverse events in resource-poor settings.

In worldwide health systems, regional anaesthesia (RA) is not applied as frequent as it should be and benefits to patient's cure and cost savings are wasted. The Regional Anaesthesia Simulator and Assistant (RASimAs) project combines image processing, physiological models, and virtual reality to support ultrasound-guided and electrical nerve stimulation-guided RA. The simulator component maps patient-specific data to general models and composes virtual reality environments using a haptic device coupled with the needle. The assistant component provides enhanced feedback mapping online-acquired ultrasound data. Regarding image processing, RASimAs aims at acquiring subject data for model development and composing a library of segmentation and registration algorithms to provide localized, patient-specific, material properties within anatomical context. Subject posing and extrapolation of body regions without patient-specific data are central challenges.

Bone age assessment (BAA) on hand radiographs is a frequent and time consuming task in radiology. We present a method for (semi)automatic BAA which is done in several steps: (i) extract 14 epiphyseal regions from the radiographs, (ii) for each region, retain image features using the IRMA framework, (iii) use these features to build a classifier model (training phase), (iv) evaluate performance on cross validation schemes (testing phase), (v) classify unknown hand images (application phase). In this paper, we combine a support vector machine (SVM) with cross-correlation to a prototype image for each class. These prototypes are obtained choosing one random hand per class. A systematic evaluation is presented comparing nominal- and real-valued SVM with k nearest neighbor (kNN) classification on 1,097 hand radiographs of 30 diagnostic classes (0 19 years). Mean error in age prediction is 1.0 and 0.83 years for 5-NN and SVM, respectively. Accuracy of nominal- and real-valued SVM based on 6 prominent regions (prototypes) is 91.57% and 96.16%, respectively, for accepting about two years age range.

Proper maintenance of sewer pipes is vital for the healthy functioning of a city. Due to the difficulty of reach for sewage pipes, automating pipe inspection has high potential in providing an efficient and objective identification of defects which could lead to damaging the pipe system. A popular approach has been to send remote controlled robots to photograph the pipes and process the images to identify possible defects. However majority of the images contain regular pipe features such as the flow line, pipe joints and pipe connections. Regular features pose a challenge for automated defect detection algorithms which require high processing time. This paper proposes a self organising map based approach to leverage the regularity of image features to isolate regions of interest which could contain defects. As a result, the search space is narrowed down for the defect detection algorithms, decreasing the overall processing time. Novelty of the work lies in the feature extraction and the gradual isolation of the potential defective image features to a manageable size. Therefore, this technique is suitable for large scale real applications. We demonstrate the effectiveness of the proposed approach for a real pipe image data set.

It is widely accepted that content-based image retrieval (CBIR) can be extremely useful for computer-aided diagnosis (CAD). However, CBIR has not been established in clinical practice yet. As a widely unattended gap of integration, a unified data concept for CBIR-based CAD results and reporting is lacking. Picture archiving and communication systems and the workflow of radiologists must be considered for successful data integration to be achieved. We suggest that CBIR systems applied to CAD should integrate their results in a picture archiving and communication systems environment such as Digital Imaging and Communications in Medicine (DICOM) structured reporting documents. A sample DICOM structured reporting template adaptable to CBIR and an appropriate integration scheme is presented. The proposed CBIR data concept may foster the promulgation of CBIR systems in clinical environments and, thereby, improve the diagnostic process.

In this paper, we present a content-based image retrieval (CBIR) system called MammoSVD. This CBIR system is developed based on breast density — fatty or dense, and the database used, from the IRMA project, provides images with the ground truth already set. Singular value decomposition (SVD) is proposed for the breast density characterization by the selection of the first singular values, in order to represent texture along with the dimensionality reduction. Support-vector machine (SVM) is used to perform the retrieval operation. Considering the first 10% of the retrieved images, the precision rate is 90%, indicating the potential of the implemented CBIR system.

One of the most significant recent advances in health information systems has been the shift from paper to electronic documents. While research on automatic text and image processing has taken separate paths, there is a growing need for joint efforts, particularly for electronic health records and biomedical literature databases. This work aims at comparing text-based versus image-based access to multimodal medical documents using state-of-the-art methods of processing text and image components. A collection of 180 medical documents containing an image accompanied by a short text describing it was divided into training and test sets. Content-based image analysis and natural language processing techniques are applied individually and combined for multimodal document analysis. The evaluation consists of an indexing task and a retrieval task based on the "gold standard" codes manually assigned to corpus documents. The performance of text-based and image-based access, as well as combined document features, is compared. Image analysis proves more adequate for both the indexing and retrieval of the images. In the indexing task, multimodal analysis outperforms both independent image and text analysis. This experiment shows that text describing images can be usefully analyzed in the framework of a hybrid text/image retrieval system.

Content--based image retrieval in the medical domain is an extremely hot topic in medical imaging as it promises to help better managing the large amount of medical images being produced. Applications are mainly expected in the field of medical teaching files and for research projects, where performance issues and speed are less critical than in the field of diagnostic aid. Final goal with most impact will be the use as a diagnostic aid in a real--world clinical setting.Other applications of image retrieval and image classification can be the automatic annotation of images with basic concepts or the control of DICOM header information.ImageCLEF is part of the Cross Language Evaluation Forum (CLEF). Since 2004, a medical image retrieval task has been added. Goal is to create databases of a realistic and useful size and also query topics that are based on real--world needs in the medical domain but still correspond to the limited capabilities of purely visual retrieval at the moment. Goal is to direct the research onto real applications and towards real clinical problems to give researchers who are not directly linked to medical facilities a possibility to work on the interesting problem of medical image retrieval based on real data sets and problems. The missing link between computer science research departments and clinical routine is one of the biggest problems that becomes evident when reading much of the current literature on medical image retrieval. Most databases are extremely small, the treated problems often far from clinical reality, and there is no integration of the prototypes into a hospital infrastructure. Only few retrieval articles specifically mention problems related to the DICOM format (Digital Imaging and Communications in Medicine) and the sheer amount of data that needs to be treated in an image archive ( > 30.000 images per day in the Geneva radiology).This article develops the various axes that can be taken into account for medical image retrieval system evaluation. First, the axes are developed based on current challenges and experiences from ImageCLEF. Then, the resources developed for ImageCLEF are listed and finally, the application of the axes is explained to show the bases of the ImageCLEFmed evaluation campaign. This article will only concentrate on the medical retrieval tasks, the non-medical tasks will only shortly be mentioned.

Medical imaging informatics must exceed the mere development of algorithms. The discipline is also responsible for the establishment of methods in clinical practice to assist physicians and improve health care. From our point of view, it is commonly accepted that model-based analysis of medical images is superior to other concepts, but only a few applications are found in daily clinical use. The gap between development of model-based image analysis and its routine application can be addressed by identifying four necessary transfer steps: formulation, parameterization, instantiation, and validation. Usually, computer scientists formulate the model and define its parameterization, i.e., configure a model to handle a selected subset of clinical data. During instantiation, the algorithm adapts the model to the actual data, which is validated by physicians. Since medical a priori knowledge and particular knowledge on technical details are required for parameterization and validation, these steps are considered to be bottlenecks. In this paper, we propose general schemes that allow an application- or image-specific parameterization to be performed by medical users. Combining noncontextual and contextual approaches, we also suggest a reliable scheme that allows application-specific validation, even if a gold standard is unavailable. To emphasize our point of view, we provide examples based on unsupervised segmentation in medical imagery, which is one of the most difficult tasks. Following the proposed schemes, an exact delineation of cells in micrographs is parameterized, validated, and successfully established in daily clinical use, while automatic determination of body regions in radiographs cannot be configured to support reliable and robust clinical use. The results stress that parameterization and validation must be based on clinical data that show all potential variations and artifact sources.

We present a discrete contour model for the segmentation of image data with any dimension of image domain and value range. The model consists of a representation using simplex meshes and a mechanical formulation of influences that drive an iterative segmentation. The object's representation as well as the influences are valid for any dimension of the image domain. The image influences introduced here, can combine information from independent channels of higher-dimensional value ranges. Additionally, the topology of the model automatically adapts to objects contained in images. Noncontextual tests have validated the ability of the model to reproducibly delineate synthetic objects. In particular, images with a signal to noise ratio of SNR /spl les/ 0.5 are delineated within two pixels of their ground truth contour. Contextual validations have shown the applicability of the model for medical image analysis in image domains of two, three, and four dimensions in single as well as multichannel value ranges.

This paper analyzes B-spline interpolation techniques of degree 2, 4, and 5 with respect to all criteria that have been applied to evaluate various interpolation schemes in a recently published survey on image interpolation in medical imaging (Lehmann et al., 1999). It is shown that high-degree B-spline interpolation has superior Fourier properties, smallest interpolation error, and reasonable computing times. Therefore, high-degree B-splines are preferable interpolators for numerous applications in medical image processing, particularly if high precision is required. If no aliasing occurs, this result neither depends on the geometric transform applied for the tests nor the actual content of images.

Local gray level dependencies of natural images can be modelled by means of co-occurrence matrices containing joint probabilities of gray-level pairs. Texture, however, is a resolution-dependent phenomenon and hence, classification depends on the chosen scale. Since there is no optimal scale for all textures we employ a multiscale approach that acquires textural features at several scales. Thus linear and nonlinear scale-spaces are analyzed by multiscale co-occurrence matrices that describe the statistical behavior of a texture in scale-space. Classification is then performed on the basis of texture features taken from the individual scale with the highest discriminatory power. By considering cross-scale occurrences of gray level pairs, the impact of filters on the feature is described and used for classification of natural textures. This novel method was found to improve classification rates of the common co-occurrence matrix approach on standard textures significantly.

Image interpolation techniques often are required in medical imaging for image generation (e.g., discrete back projection for inverse Radon transform) and processing such as compression or resampling. Since the ideal interpolation function spatially is unlimited, several interpolation kernels of finite size have been introduced. This paper compares 1) truncated and windowed sinc; 2) nearest neighbor; 3) linear; 4) quadratic; 5) cubic B-spline; 6) cubic; g) Lagrange; and 7) Gaussian interpolation and approximation techniques with kernel sizes from 1 x 1 up to 8 x 8. The comparison is done by: 1) spatial and Fourier analyses; 2) computational complexity as well as runtime evaluations; and 3) qualitative and quantitative interpolation error determinations for particular interpolation tasks which were taken from common situations in medical image processing. For local and Fourier analyses, a standardized notation is introduced and fundamental properties of interpolators are derived. Successful methods should be direct current (DC)-constant and interpolators rather than DC-inconstant or approximators. Each method's parameters are tuned with respect to those properties. This results in three novel kernels, which are introduced in this paper and proven to be within the best choices for medical image interpolation: the 6 x 6 Blackman-Harris windowed sinc interpolator, and the C2-continuous cubic kernels with N = 6 and N = 8 supporting points. For quantitative error evaluations, a set of 50 direct digital X rays was used. They have been selected arbitrarily from clinical routine. In general, large kernel sizes were found to be superior to small interpolation masks. Except for truncated sinc interpolators, all kernels with N = 6 or larger sizes perform significantly better than N = 2 or N = 3 point methods (p << 0.005). However, the differences within the group of large-sized kernels were not significant. Summarizing the results, the cubic 6 x 6 interpolator with continuous second derivatives, as defined in (24), can be recommended for most common interpolation tasks. It appears to be the fastest six-point kernel to implement computationally. It provides eminent local and Fourier properties, is easy to implement, and has only small errors. The same characteristics apply to B-spline interpolation, but the 6 x 6 cubic avoids the intrinsic border effects produced by the B-spline technique. However, the goal of this study was not to determine an overall best method, but to present a comprehensive catalogue of methods in a uniform terminology, to define general properties and requirements of local techniques, and to enable the reader to select that method which is optimal for his specific application in medical imaging.

Digital subtraction has shown to be the best noninvasive tool for diagnosing small changes of internal structures in medical images. However, subtraction techniques require perfect geometric match. Geometric projection of most medical imaging techniques could be modeled by perspective projection. Eight parameters must be determined for the perspective back-projection of the subsequent image into the geometry of the reference. Perspective invariants which allow the determination of the total eight parameters are still unknown. Therefore, rotations, translations, and scalings (RST-movements) are detected in a first stage using the Fourier-Mellin invariant. This pre-registration allows the linkage of corresponding landmarks in both images by local correlation. In the second stage, the coordinates of automatically detected landmarks are used to determine the perspective parameters for fine-adjustment. The two stages form a robust registration algorithm which is applied to the alignment of intraoral radiographs acquired without individual adjustment aids. The excellent results allow the clinical establishment of the method.

In order to measure the angle of squint, the eyes of a patient are photographed using two CCD-cameras. Three hashes are activated at the moment of exposure. By measuring the positions of the reflexes in the digital image the squint angle can be determined. This paper presents a segmentation method for the automatic detection of the reflexes without the use of fixed threshold values. With an algorithm based on Hough-transformation the centre of the iris is localized. In the region of interest the reflex positions are located using cross-covariance-filtering. A classification and plausibility control follow to make sure there are no faulty measurements. With 191 clinical images arbitrarily chosen the algorithm was tested showing perfect results. It is already used in clinical practice.>

Describes the image processing method for the automatisation of the measurement of squint using the first and fourth Purkinje reflex. This method is already in use in clinical practice in comparison to the standard method of the cover test. The new method, based on image processing, allows the exact and objective determination of the angle of squint. Children from the age of six months as well as adults can be examined. The automatisation of the previously interactive method is a prerequisite for the screening of large collectives of children for microstrabism. The method uses a two-stage abstraction to identify six reflexes in the iris of each eye. The method based on the Hough-transform and covariance-filtering is robust against image distortions and additional reflexes, as produced by wearing glasses. The reliability of the method has been proven in 191 cases and is tested in routine use now.