Two models for outcome prediction - a comparison of logistic regression and neural networks.

OBJECTIVES: Accurately predicting disease progress from a set of predictive variables is an important aspect of clinical work. For binary outcomes, the classical approach is to develop prognostic logistic regression (LR) models. Alternatively, machine learning algorithms were proposed with artificial neural networks (ANN) having become popular over the last decades. Although some studies have compared predictive accuracies of LR and ANN models, some concerns regarding their methodological quality have been voiced. Our comparison has the advantage of being based on two large independent data sets allowing for elaborate model development and independent validation. METHODS: From the German Stroke Database, a learning data set including 1754 prospectively recruited patients with acute ischemic stroke was used. Utilizing LR and ANN, two prognostic models were developed predicting restitution of functional independence and survival after 100 days. The resulting models were applied to classify 1470 patients with acute ischemic stroke; this test data set was collected independently from the learning data. Error fractions in the test data were determined, and differences in error fractions between the algorithms were calculated with 95% confidence intervals. RESULTS: For most prognostic models, error fractions in the test data were below 40%. There was no difference between the algorithms except for the model predicting completely versus incompletely restituted or deceased patients (difference in error fractions = 4.01% [2.10-5.96%], p = 0.0001). CONCLUSIONS: The conscientiously applied LR remains the gold standard for prognostic modelling; however, ANN can be an alternative automated "quick and easy" multivariate analysis.

Atlas-based recognition of anatomical structures and landmarks and the automatic computation of orthopedic parameters.

OBJECTIVE: This paper describes methods for the automatic atlas-based segmentation of bone structures of the hip, the automatic detection of anatomical point landmarks and the computation of orthopedic parameters to avoid the interactive, time-consuming pre-processing steps for the virtual planning of hip operations. METHODS: Based on the CT data of the Visible Human Data Sets, two three-dimensional atlases of the human pelvis have been built. The atlases consist of labeled CT data sets, 3D surface models of the separated structures and associated anatomical point landmarks. The atlas information is transferred to the patient data by a non-linear gray value-based registration algorithm. A surface-based registration algorithm was developed to detect the anatomical landmarks on the patient's bone structures. Furthermore, a software tool for the automatic computation of orthopedic parameters is presented. Finally, methods for an evaluation of the atlas-based segmentation and the atlas-based landmark detection are explained. RESULTS: A first evaluation of the presented atlas-based segmentation method shows the correct labeling of 98.5% of the bony voxels. The presented landmark detection algorithm enables the precise and reliable localization of orthopedic landmarks. The accuracy of the landmark detection is below 2.5 mm. CONCLUSION: The atlas-based segmentation of bone structures, the atlas-based landmark detection and the automatic computation of orthopedic measures are suitable to essentially reduce the time-consuming user interaction during the pre-processing of the CT data for the virtual three-dimensional planning of hip operations.

The capabilities of artificial neural networks in body composition research.

When estimating in vivo body composition or combining such estimates with other results, multiple variables must be taken into account (e. g. binary attributes such as gender or continuous attributes such as most biosignals). Standard statistical models, such as logistic regression and multivariate analysis, presume well-defined distributions (e. g. normal distribution); they also presume independence among all inputs and only linear relationships, yet rarely are these requirements met in real life. As an alternative to these models, artificial neural networks can be used. In the present work, we describe the pre-processing and multivariate analysis of data using neural network techniques, providing examples from the medical field and making comparisons with classic statistical approaches. We also address the criticisms raised regarding neural network techniques and discuss their potential improvement.

Predicting the intracellular water compartment using artificial neural network analysis.

Artificial neural networks (ANN) are used for a wide variety of data-processing applications such as predicting medical outcomes and classifying clinical data and patients. We investigated the applicability of an ANN for estimating the intracellular water compartment for a population of 104 healthy Italians ranging in age from 19 to 68 years. Anthropometric variables, bioelectric impedance analysis (BIA) variables, and reference values for intracellular water, measured using whole-body (40)K counting (ICW(K40)), were measured for all study participants. The anthropometric variables and the impedance index (height(2)/resistance) were fed to the ANN input layer, which produced as output the estimated values for intracellular water (ICW(ANN)). We also estimated intracellular water using a BIA formula for the same population (ICW(DeLorenzo)) and another for Caucasians (ICW(Gudivaka)). Errors in the estimations generated by ANN and the BIA equations were calculated as the root mean square error (RMSE). The mean (+/-SD) reference value (ICWK40) was 25.01+/-4.50 l, whereas the mean estimated value was 15.20+/-1.79 l (RMSE=11.06 l) when calculated using ICW(DeLorenzo), 18.07+/-1.14 l (RMSE=8.72 l) when using ICW(Gudivaka), and 25.01+/-2.74 l (RMSE=3.22 l) when using ICW(ANN). Based on these results, we deduce that the ANN algorithm is a more accurate predictor for reference ICW(K40) than BIA equations.

Artificial neural network analysis: a novel application for predicting site-specific bone mineral density.

Dual X-ray absorptiometry (DXA), which is the most commonly used method for the diagnosis and followup of human bone health, is known to produce accurate estimates of bone mineral density (BMD). However, high costs and problems with availability may prevent its use for mass screening. The objective of the present study was to estimate BMD values for healthy persons and those with conditions known to be associated with BMD, using artificial neural networks (ANN). An ANN was used to quantitatively estimate site-specific BMD values in comparison with reference values obtained by DXA (i. e. BMD(spine), BMD(pelvis), and BMD(total)). Anthropometric measurements (i. e. sex, age, weight, height, body mass index, waist-to-hip ratio, and the sum of four skinfold thicknesses) were fed to the ANN as independent input variables. The estimates based on four input variables were generated as output and were generally identical to the reference values for all studied groups. We believe the ANN is a promising approach for estimating and predicting site-specific BMD values using simple anthropometric measurements.

Predicting Type 2 diabetes using an electronic nose-based artificial neural network analysis.

Diabetes is a major health problem in both industrial and developing countries, and its incidence is rising. Although detection of diabetes is improving, about half of the patients with Type 2 diabetes are undiagnosed and the delay from disease onset to diagnosis may exceed 10 yr. Thus, earlier detection of Type 2 diabetes and treatment of hyperglycaemia and related metabolic abnormalities is of vital importance. The objectives of the present study were to examine urine samples from Type 2 diabetic patients and healthy volunteers using the electronic nose technology and to evaluate possible application of data classification methods such as self-learning artificial neural networks (ANN) and logistic regression (LR) in comparison with principal components analysis (PCA). Urine samples from Type 2 diabetic patients and healthy controls were processed randomly using a simple 8-sensors electronic nose and individual electronic nose patterns were qualitatively classified using the "Approximation and Classification of Medical Data" (ACMD) network based on 2 output neurons, binary LR analysis and PCA. Distinct classes were found for Type 2 diabetic subjects and controls using PCA, which had a 96.0% successful classification percentage mean while qualitative ANN analysis and LR analysis had successful classification percentages of 92.0% and 88.0%, respectively. Therefore, the ACMD network is suitable for classifying medical and clinical data.

Atlas-based segmentation of bone structures to support the virtual planning of hip operations.

Two 3-D digitised atlases of a female and a male pelvis were generated to support the virtual 3-D planning of hip operations. The anatomical atlases were designed to replace the interactive, time-consuming pre-processing steps for the virtual operation planning. Each atlas consists of a labelled reference CT data set and a set of anatomical point landmarks. The paper presents methods for the automatic transfer of these anatomical labels to an individual patient data set. The labelled patient data are used to generate 3-D models of the patient's bone structures. Besides the anatomical labelling, the determination of measures, like angles, distances or sizes of contact areas, is important for the planning of hip operations. Thus, algorithms for the automatic computation of orthopaedic parameters were implemented. A first evaluation of the presented atlas-based segmentation method shows a correct labelling of 98.5% of the bony voxels.

Three-dimensional planning and simulation of hip operations and computer-assisted construction of endoprostheses in bone tumor surgery.

OBJECTIVE: This article presents the VIRTOPS (VIRTual Operation Planning in Orthopaedic Surgery) software system for virtual preoperative planning and simulation of hip operations. The system is applied to simulate the endoprosthetic reconstruction of the hip joint with hemipelvic replacement, and supports the individual design of anatomically adaptable, modular prostheses in bone tumor surgery. The virtual planning of the operation and the construction of the individual implant are supported by virtual reality techniques. The central step of the operation planning procedure, the placement of the cutting plane in the hip bone, depends strongly on the tumor's position. Segmentation of the tumor and the bones in MR and CT data, as well as fusion of MR and CT image sequences, is necessary to visualize the tumor's position within the hip bone. MATERIALS AND METHODS: Three-dimensional models of the patient's hip are generated based on CT image data. A ROI-based segmentation algorithm enables the separation of the bone tumor in multispectral MR image sequences. A special registration method using segmentation results has been developed to transfer CT and MR data into one common coordinate system. During the 3D planning process, the surgeon simulates the operation and defines the position and geometry of the custom-made endoprosthesis. Stereoscopic visualization and 3D input devices facilitate navigation and 3D interaction in the virtual environment. Special visualization techniques such as texture mapping, color coding of quantitative parameters, and transparency support the determination of the correct position and geometry of the prosthesis. RESULTS AND CONCLUSIONS: The VIRTOPS system enables the complete virtual planning of hip operations with endoprosthetic reconstruction, as well as the optimal placement and design of endoprostheses. After the registration and segmentation of CT and MR data, 3D visualizations of the tumor within the bone are generated to support the surgeon during the planning procedure. In the virtual planning environment, individually adapted endoprostheses can be constructed without the need to generate expensive solid 3D models. Furthermore, different operation strategies can be compared easily. Three-dimensional images and digital movies generated during the virtual operation planning can be used for case documentation and patient information purposes.

Simulation of hip operations and design of custom-made endoprostheses using virtual reality techniques.

In this paper a system for the virtual planning of hip operations with endoprosthetic reconstruction and its application in bone tumor surgery is described. The system enables the simulation of the operation and the construction of a custom-made implant depending on the chosen resection planes and the patient's anatomy. During the planning process integrated virtual reality techniques facilitate the interaction with the three-dimensional (3D) medical objects. Stereo viewing improves the perception of the 3D nature of bone structures and tumors. In comparison to conventional planning procedures, different operation strategies and their influence on the geometry of the custom-made endoprosthesis can be easily compared. Furthermore, the combination of multi-modal image information (CT and MR) enables an accurate 3D visualization of the bone tumor within the bone.

Virtual planning of hip operations and individual adaption of endoprostheses in orthopaedic surgery.

The introduction of virtual reality techniques in medicine opens up new possibilities for the planning of interventions. The presented software system for virtual operation planning in orthopaedic surgery (VIRTOPS) enables the virtual preoperative 3D planning and simulation of pelvis and hip operations. It is used to plan operations of bone tumours with endoprosthetic reconstruction of the hip based on multimodal image information. The operation and the endosprothetic reconstruction of the pelvis are simulated using virtual reality techniques. Stereoscopic visualisation techniques and 3D input devices support the 3D interaction with the virtual 3D models. The main task of the preoperative planning process is the individual design of an anatomically adaptable modular prosthesis. The placement and the design of the endoprosthesis are supported by different functions and visualisation techniques. The resulting 3D images and movies can be used for the documentation of the operation planning procedure, as well as, for the preoperative information of the patient.

Artificial neural networks for classifying olfactory signals.

For practical applications, artificial neural networks have to meet several requirements: Mainly they should learn quick, classify accurate and behave robust. Programs should be user-friendly and should not need the presence of an expert for fine tuning diverse learning parameters. The present paper demonstrates an approach using an oversized network topology, adaptive propagation (APROP), a modified error function, and averaging outputs of four networks described for the first time. As an example, signals from different semiconductor gas sensors of an electronic nose were classified. The electronic nose smelt different types of edible oil with extremely different a-priori-probabilities. The fully-specified neural network classifier fulfilled the above mentioned demands. The new approach will be helpful not only for classifying olfactory signals automatically but also in many other fields in medicine, e.g. in data mining from medical databases.

An anatomical atlas to support the virtual planning of hip operations.

Two three-dimensional digitized atlases of a female and a male pelvis were generated to support the virtual 3D-planning of hip operations. Beside the anatomical labeling of bone structures the determination of orthopedic measures, like angles, distances or sizes of contact areas, is important for the planning of hip operations. Thus, each atlas consists of labeled reference CT data sets, a set of landmarks as well as definitions of orthopedic measures and methods for their automatic computation. Furthermore, methods for the automatic transfer of anatomical labels from the atlas to an individual data set are presented resulting in three-dimensional models of the patient's bone structures. The anatomical atlases are designed to replace the interactive, time-consuming pre-processing steps for the virtual 3D operation planning.

Feature selection for optimized skin tumor recognition using genetic algorithms.

In this paper, a new approach to computer supported diagnosis of skin tumors in dermatology is presented. High resolution skin surface profiles are analyzed to recognize malignant melanomas and nevocytic nevi (moles), automatically. In the first step, several types of features are extracted by 2D image analysis methods characterizing the structure of skin surface profiles: texture features based on cooccurrence matrices, Fourier features and fractal features. Then, feature selection algorithms are applied to determine suitable feature subsets for the recognition process. Feature selection is described as an optimization problem and several approaches including heuristic strategies, greedy and genetic algorithms are compared. As quality measure for feature subsets, the classification rate of the nearest neighbor classifier computed with the leaving-one-out method is used. Genetic algorithms show the best results. Finally, neural networks with error back-propagation as learning paradigm are trained using the selected feature sets. Different network topologies, learning parameters and pruning algorithms are investigated to optimize the classification performance of the neural classifiers. With the optimized recognition system a classification performance of 97.7% is achieved.

Computer-supported diagnosis of melanoma in profilometry.

Laser profilometry offers new possibilities to improve non-invasive tumor diagnostics in dermatology. In this paper, a new approach to computer-supported analysis and interpretation of high-resolution skin-surface profiles of melanomas and nevocellular nevi is presented. Image analysis methods are used to describe the profile's structures by texture parameters based on co-occurrence matrices, features extracted from the Fourier power spectrum, and fractal features. Different feature selection strategies, including genetic algorithms, are applied to determine the best possible subsets of features for the classification task. Several architectures of multilayer perceptrons with error back-propagation as learning paradigm are trained for the automatic recognition of melanomas and nevi. Furthermore, network-pruning algorithms are applied to optimize the network topology. In the study, the best neural classifier showed an error rate of 4.5% and was obtained after network pruning. The smallest error rate in all, of 2.3%, was achieved with nearest neighbor classification.

Computer-assisted planning and simulation of hip operations using virtual three-dimensional models.

In this paper a new software system for virtual preoperative planning and simulation of hip operations is presented. The system simulates the endoprosthetic reconstruction of the hip joint with hemipelvic replacement for bone tumor patients and supports the individual design of anatomically adaptable, modular prostheses. Three-dimensional models of the patient's hip are generated based on CT data. The surgeon simulates the operation and defines the position and shape of the custom-made endoprosthesis. Stereoscopic visualization and 3D input devices facilitate the navigation and interaction in the virtual environment. Special visualization techniques like texture mapping, color coding of quantitative parameters or transparency support the determination of the correct position and shape of the prosthesis. Furthermore, the system can be used for patient information or educational tasks.

Image analysis and pattern recognition for computer supported skin tumor diagnosis.

A new approach to computer supported recognition of melanoma and naevocytic naevi based on high resolution skin surface profiles is presented. Profiles are generated by sampling an area of 4 x 4 mm2 at a resolution of 125 sample points per mm with a laser profilometer at a vertical resolution of 0.1 micron. With image analysis algorithms Haralick's texture parameters, Fourier features and features based on fractal analysis are extracted. Genetic algorithms are employed successfully to select good feature subsets for the following classification process. As quality measure for feature subsets, the error rate of the nearest neighbor classifier estimated with the leaving-one-out method is used. Classification is performed with feed forward back-propagation network and the nearest neighbor classifier. Classification performance of the neural classifier is optimized using different topologies, learning parameters and pruning algorithms. The best neural classifier achieved an error rate of 4.5% and was found after network pruning. The best result with an error rate of 2.3% was obtained with the nearest neighbor classifier.

[Future of endoscopic surgery--German health law and minimal access surgery in an university hospital]. / Zukunft der endoskopischen Chirurgie--GSG-Probleme der MIC in einer Universitätsklinik.

The future development of endoscopic surgery depends on a medical and economical benefit. Medical advantages are demonstrated under professional conditions of practice in a retrospective study: endoscopical repair of the groin (TEP n = 44) vs. conventional Shouldice- (n = 17) and Lichtenstein (n = 19) method, laparoscopical hemifundoplication (n = 7) vs. traditional Nissen-Rosetti procedure (n = 3) and also resection of the sigmoid (lap. n = 26) vs. open surgery (n = 12). The overall hospital stay is shortend dramatically (primary hernia of the groin 8.8 (Shouldice) and 7.4 (Lichtenstein) vs. 3.1 days (TEP); (hemi-) fundoplication 11.1 (open) vs. 5.0 days (lap.); resection of sigmoid 19.0 (open) vs.17.0 days (lap.)) At the same time quality of care is held or improved. Comparison of real cost analysis revealed a better economical result (593-970 DM lower cost for TEP, 1.256 DM lower costs for lap. hemifundoplication, and 1.918 DM in case of lap. resection of sigmoid) for minimal-access-surgery (MAS), although particular costs for the endoscopic surgical procedure are increasing up to 100%, especially at the beginning (learning curve). The German payment-system does not regard the special conditions of MAS. There is no case-related payment for MAS due to the lower overall costs. Therefore the financial result is worse than for conventional treatment.

KAMEDIN: a telemedicine system for computer supported cooperative work and remote image analysis in radiology.

The software system KAMEDIN (Kooperatives Arbeiten und MEdizinische Diagnostik auf Innovativen Netzen) is a multimedia telemedicine system for exchange, cooperative diagnostics, and remote analysis of digital medical image data. It provides components for visualisation, processing, and synchronised audio-visual discussion of medical images. Techniques of computer supported cooperative work (CSCW) synchronise user interactions during a teleconference. Visibility of both local and remote cursor on the conference workstations facilitates telepointing and reinforces the conference partner's telepresence. Audio communication during teleconferences is supported by an integrated audio component. Furthermore, brain tissue segmentation with artificial neural networks can be performed on an external supercomputer as a remote image analysis procedure. KAMEDIN is designed as a low cost CSCW tool for ISDN based telecommunication. However it can be used on any TCP/IP supporting network. In a field test, KAMEDIN was installed in 15 clinics and medical departments to validate the systems' usability. The telemedicine system KAMEDIN has been developed, tested, and evaluated within a research project sponsored by German Telekom.

Comparative coherence studies in healthy volunteers and Down's syndrome patients from childhood to adult age.

Within the scope of the Munich Pediatric Longitudinal Study, EEG coherence was studied in 212 Down's syndrome patients and 342 healthy controls aged from 6 months up to 30 years. The digitalized EEG records were subjected to spectral analysis. Frequency band-related coherences were calculated to reveal age-specific differences in the functional relationship between two brain areas in Down's syndrome patients and controls. The results show that in the "eyes-open" state the intra-hemispheric coherence in the alpha band was significantly lower (P less than 0.05) in the Down's syndrome patients than in the controls whereas that in the delta bands it was generally higher. The intra-hemispheric coherence in the "eyes-closed" state was generally higher in the Down's syndrome groups than in the controls; however, significant differences could be detected only in some age groups. The age-specific development of coherence in the inter-hemispheric parieto-occipital region was almost identical in Down's syndrome children as in controls, both with open and closed eyes. The most distinct differences were found in the fronto-central inter-hemispheric coherence (P less than 0.01), while the coherence deficiencies in the Down's syndrome group became more prominent with increasing age from school age onwards. These electrophysiological results are compared with the results of neuropathological and neurophysiological studies of other authors. It can be suggested that there are correlations with a significantly small number of dendritic spines in Down's syndrome patients, which was determined in neuropathological examinations. A neuronal model of interpretation is presented which explains the increasing developmental deficit with age in Down's syndrome children.