RESUMO
BACKGROUND: Contrast-enhanced ultrasound (CEUS) is considered as an efficient tool for focal liver lesion characterization, given it allows real-time scanning and provides dynamic tissue perfusion information. An accurate diagnosis of liver lesions with CEUS requires a precise interpretation of CEUS images. However,it is a highly experience dependent task which requires amount of training and practice. To help improve the constrains, this study aims to develop an end-to-end method based on deep learning to make malignancy diagnosis of liver lesions using CEUS. METHODS: A total of 420 focal liver lesions with 136 benign cases and 284 malignant cases were included. A deep learning model based on a two-dimensional convolution neural network, a long short-term memory (LSTM), and a linear classifier (with sigmoid) was developed to analyze the CEUS loops from different contrast imaging phases. For comparison, a 3D-CNN based method and a machine-learning (ML)-based time-intensity curve (TIC) method were also implemented for performance evaluation. RESULTS: Results of the 4-fold validation demonstrate that the mean AUC is 0.91, 0.88, and 0.78 for the proposed method, the 3D-CNN based method, and the ML-based TIC method, respectively. CONCLUSIONS: The proposed CNN-LSTM method is promising in making malignancy diagnosis of liver lesions in CEUS without any additional manual features selection.
Assuntos
Aprendizado Profundo , Neoplasias Hepáticas , Humanos , Neoplasias Hepáticas/diagnóstico por imagem , Neoplasias Hepáticas/patologia , Meios de Contraste , Ultrassonografia/métodosRESUMO
BACKGROUND: This study was performed to develop and validate machine learning models for early detection of ventilator-associated pneumonia (VAP) 24 h before diagnosis, so that VAP patients can receive early intervention and reduce the occurrence of complications. PATIENTS AND METHODS: This study was based on the MIMIC-III dataset, which was a retrospective cohort. The random forest algorithm was applied to construct a base classifier, and the area under the receiver operating characteristic curve (AUC), sensitivity and specificity of the prediction model were evaluated. Furthermore, We also compare the performance of Clinical Pulmonary Infection Score (CPIS)-based model (threshold value ≥ 3) using the same training and test data sets. RESULTS: In total, 38,515 ventilation sessions occurred in 61,532 ICU admissions. VAP occurred in 212 of these sessions. We incorporated 42 VAP risk factors at admission and routinely measured the vital characteristics and laboratory results. Five-fold cross-validation was performed to evaluate the model performance, and the model achieved an AUC of 84% in the validation, 74% sensitivity and 71% specificity 24 h after intubation. The AUC of our VAP machine learning model is nearly 25% higher than the CPIS model, and the sensitivity and specificity were also improved by almost 14% and 15%, respectively. CONCLUSIONS: We developed and internally validated an automated model for VAP prediction using the MIMIC-III cohort. The VAP prediction model achieved high performance based on its AUC, sensitivity and specificity, and its performance was superior to that of the CPIS model. External validation and prospective interventional or outcome studies using this prediction model are envisioned as future work.
Assuntos
Pneumonia Associada à Ventilação Mecânica , Cuidados Críticos , Humanos , Unidades de Terapia Intensiva , Aprendizado de Máquina , Pneumonia Associada à Ventilação Mecânica/diagnóstico , Pneumonia Associada à Ventilação Mecânica/epidemiologia , Estudos Prospectivos , Estudos RetrospectivosRESUMO
Sensitive band positions, models and the principles of soil dispersion detected by hyperspectral remote sensing were firstly discussed according to the results of soil dispersive hyperspectral remote sensing experiment. Results showed that, (1) signals and noises could be separated by Fourier transformation. A finely mineral identification system was developed to remove spectral noises and provide highly accurate data for establishing soil dispersive model; (2) Soil dispersive hyperspectral remote sensing model established by the multiple linear regression method was good at soil dispersion forecasting for the high correlation between sensitive bands and the soil dispersions. (3) According to mineral spectra, soil minerals and their absorbed irons were reflected by sensitive bands which revealed reasons causing soils to be dispersive. Sodium was the closest iron correlated with soil dispersion. The secondary was calcite, montmorillonite and illite. However, the correlation between soil dispersion and chlorite, kaolinite, PH value, quartz, potassium feldspar, plagioclase was weak. The main reason was probably that sodium was low in ionic valence, small ionic radius and strong hydration forces; calcite was high water soluble and illite was weak binding forces between two layers under high pH value.
RESUMO
To improve the accuracy of mineral content extraction by linear decomposition model, a method was established, which took rock spectra with wavelength from 350 to 2 500 nm as the data source, identified minerals based on spectral matching methods, applied Hapke model to transform spectral reflectance into single scattering albedo and resolved single scattering albedo to get mineral content. In this method, sectional noise filtering and regional mineral spectra library were added to improve the identifying accuracy. Based on the analysis on the fifth Baogutu rock body, compared with XRD results, accuracies of quartz, feldspar class and altered minerals identification were 75%, 100% and 92.2% separately. Accuracy of the content extraction of feldspar class, hornblende and altered minerals were 80.5%, 64%, 92.36% separately. This method added mineralogy symbiotic relationship into mineral identification to ensure the reliability, proposed the idea of sectional noise filtering to avoid the influence of filtering algorithm, applied the single scattering albedo to avoid the complex nonlinearly calculations to improve the accuracy theoretically. This method has a certain guiding significance for the work such as rapid analysis of alteration information.
RESUMO
Model selection for support vector machine (SVM) involving kernel and the margin parameter values selection is usually time-consuming, impacts training efficiency of SVM model and final classification accuracies of SVM hyperspectral remote sensing image classifier greatly. Firstly, based on combinatorial optimization theory and cross-validation method, artificial immune clonal selection algorithm is introduced to the optimal selection of SVM (CSSVM) kernel parameter a and margin parameter C to improve the training efficiency of SVM model. Then an experiment of classifying AVIRIS in India Pine site of USA was performed for testing the novel CSSVM, as well as a traditional SVM classifier with general Grid Searching cross-validation method (GSSVM) for comparison. And then, evaluation indexes including SVM model training time, classification overall accuracy (OA) and Kappa index of both CSSVM and GSSVM were all analyzed quantitatively. It is demonstrated that OA of CSSVM on test samples and whole image are 85.1% and 81.58, the differences from that of GSSVM are both within 0.08% respectively; And Kappa indexes reach 0.8213 and 0.7728, the differences from that of GSSVM are both within 0.001; While the ratio of model training time of CSSVM and GSSVM is between 1/6 and 1/10. Therefore, CSSVM is fast and accurate algorithm for hyperspectral image classification and is superior to GSSVM.
RESUMO
Rapid identification of minerals based on near infrared (NIR) and shortwave infrared (SWIR) hyperspectra is vital to remote sensing mine exploration, remote sensing minerals mapping and field geological documentation of drill core, and have leaded to many identification methods including spectral angle mapping (SAM), spectral distance mapping (SDM), spectral feature fitting(SFF), linear spectral mixture model (LSMM), mathematical combination feature spectral linear inversion model(CFSLIM) etc. However, limitations of these methods affect their actual applications. The present paper firstly gives a unified minerals components spectral inversion (MCSI) model based on target sample spectrum and standard endmember spectral library evaluated by spectral similarity indexes. Then taking LSMM and SAM evaluation index for example, a specific formulation of unified MCSI model is presented in the form of a kind of combinatorial optimization. And then, an artificial immune colonial selection algorithm is used for solving minerals feature spectral linear inversion model optimization problem, which is named ICSFSLIM. Finally, an experiment was performed to use ICSFSLIM and CFSLIM to identify the contained minerals of 22 rock samples selected in Baogutu in Xinjiang China. The mean value of correctness and validness identification of ICSFSLIM are 34.22% and 54.08% respectively, which is better than that of CFSLIM 31.97% and 37.38%; the correctness and validness variance of ICSFSLIM are 0.11 and 0.13 smaller than that of CFSLIM, 0.15 and 0.25, indicating better identification stability.
RESUMO
Aiming at the low accuracy of mineral identification with hyperspectral data, the present article established regional spectra library on the basis of the study area geological background, and presented a pretreatment method that filters the original spectra by section. First, continuum based fast Fourier transform was used to filter the noise among 2000-2200, 2250-2300 and 2350-2500 nm. Then apply the rapid quantificational identification model with regional spectrum library was used to dispose the processed spectra. The highest effective rate of the result is 80%, and the highest accuracy rate is 67%. Compared with the identification result of original spectra, the average accuracy rate was upgraded by 17.7%, and the average effective rate was upgraded by 5.1%. Compared with the identification result of all-filtered spectra, the average accuracy rate was upgraded by 5.8%, while the average effective rate was upgraded by 39.8%. This method, which could guarantee that the identification result contains the most correct minerals and the fewest error ones, promoted mineral identification accuracy. The result with higher accuracy is significant to rapid mineral extraction work in field.
RESUMO
Geological section can help validating and understanding of the alteration information which is extracted from remote sensing images. In the paper, the concept of spectral geological profile was introduced based on the principle of geological section and the method of spectral information extraction. The spectral profile can realize the storage and vision of spectra along the geological profile, but the spectral geological spectral profile includes more information besides the information of spectral profile. The main object of spectral geological spectral profile is to obtain the distribution of alteration types and content of minerals along the profile which can be extracted from spectra measured by field spectrometer, especially for the spatial distribution and mode of alteration association. Technical method and work flow of alteration information extraction was studied for the spectral geological profile. The spectral geological profile was set up using the ground reflectance spectra and the alteration information was extracted from the remote sensing image with the help of typical spectra geological profile. At last the meaning and effect of the spectral geological profile was discussed.
RESUMO
The rapid identification of the minerals in the field is crucial in the remote sensing geology study and mineral exploration. The characteristic spectrum linear inversion modeling is able to obtain the mineral information quickly in the field study. However, the authors found that there was significant difference among the results of the model using the different kinds of spectra of the same sample. The present paper mainly studied the continuum based fast Fourier transform processing (CFFT) method and the characteristic spectrum linear inversion modeling (CSLM). On one hand, the authors obtained the optimal preferences of the CFFT method when applying it to rock samples: setting the CFFT low-pass frequency to 150 Hz. On the other hand, through the evaluation and study of the results of CSLM using different spectra, the authors found that the ASD spectra which were denoised in the CFFT method could provide better results when using them to extract the mineral information in the field.
RESUMO
In order to explore the feasibility of studying the geochemical anomaly of copper element by using remote sensing method, the correlation between Cu and other elements and the correlation between Cu and reflectance spectra were analyzed based on the element contents and the reflectance spectra of rock samples. It was found that Fe is most highly correlated with Cu, followed by Ti and As. The relationship between the Cu content and the reflectance spectra is of a negative correlation, and the higher the Cu content, the stronger the correlation. Furthermore, based on the reflectance spectra, the partial least squares regression of the Cu, Fe, Ti and As content was carried out respectively. The result shows that Ti gets the highest accuracy, followed by Fe. The worst is for As. Although the accuracy of the Cu model is not too high, it is feasible to establish an indirect model of copper anomaly on the basis of Fe model because of the strong correlation between them. In order to improve the accuracy of the model, some transformations for the reflectance spectra were performed and many spectral indices were acquired. Based on the spectral indices, the partial least squares regression of Fe was carried out. The accuracy of the regression model increased greatly. The highest correlation coefficient of the regression model is 0.687 6 for the calibration samples and it is 0.595 9 for the validation samples.
RESUMO
The present study introduced the generalized morphological filter into the denoising of visible and near infrared spectra for the first time, and provided a new method for denoising the reflectance spectra by combining mathematical morphology methods with the wavelet packet transformation. The authors used vegetable spectra from USGS spectral library as the reference spectra, and obtained the noised spectra by adding noises with different signal-to-noise ratios to the referenced spectra. The results were evaluated by signal-to-noise ratio (SNR), root mean squared error (RMSE), normalized correlation coefficient (NCC) and smoothness ratio (SR) of the denoised spectra. The authors' results showed that both the thresholding on wavelet packet decomposition best bases method and the generalized morphological filter method could maintain the spectral shape and the spectral smoothness after denoising. The generalized morphological filter method can remove larger amplitude random noise whereas the continuous small amplitude random noise could not be removed well. Hence, the denoised spectra were not smooth. Nevertheless, the denoised spectra using the thresholding on the best base groups of wavelet packet decomposition method were smooth, but the larger amplitude noise could not be removed completely. The authors' method by combining the two methods has the merits of the two methods but removing their defects. The results showed that both large and small amplitude noise could be removed completely, meanwhile the normalized correlation coefficient (NCC) and smoothness ratio (SR) were improved, which indicated that the authors' method is superior to other methods in denoising visible and near infrared spectra.
RESUMO
The present paper presents a new alteration mineral mapping method based on statistical analysis of field measured spectra. First of all, this method processes a cluster of measurement data of spectra of field samples, in order to distinguish different sample area from the overall types. Second, the results of the clustering of different mineral alterations established their respective discriminant functions. Thus, mapping major alteration type accords with the clustered reference spectra by given remote sensing images. Finally mapping further alteration types was based on the discriminant function of second step, which leads to final alteration map. This method takes full account of the different combination of alteration types, as well as the regional differences of alterations, and the establishment of the discriminant function for alteration minerals is more scientific. Moreover, the authors accessed the reliability of mapping to a certain extent. The method was applied to a study area of Baogutu in Xinjiang Province, which represents a good result.
RESUMO
Hyperspectral characteristics analysis of ground features is the basis for applications of high-resolution imaging technology to ground target identification and ground features classification. Based on morphological multi-scale Top-Hat transformation, a novel spectral absorption enhancing algorithms was put forward, which enhanced spectral absorption features while maintaining shape features of the absorption peak bands. Eleven reflectance spectra of different mineral groups were chosen from the mineral spectral library of the United States Geological Survey (USGS), and we used a K-means clustering analysis on both the absorption-enhanced spectra and the original reflectance spectra. Results showed that, firstly, clustering groups of the absorption-enhanced spectra (AES) had better similarity within the same clustering group, and greater difference between different groups, furthermore, they were more consistent with the geological background of these minerals compared with clustering result of the original spectra (OS). Secondly, while all the original spectra were re-sampled to their ASTER spectra and the AES clustering result was displayed in the form of ASTER spectra of the minerals, we could easily describe both the representative spectral feature of each clustering group, and the typical spectral differences between every two groups. These fully demonstrate that the absorption-enhanced spectra have enhanced absorption features of the mineral spectra, and improved the separability of hyper-spectra. Accordingly, feature analysis based on absorption enhanced spectra can be used as reference for information extracting based on multi-spectral remote sensing image data, and it is a very useful method of hyperspectral analysis.
RESUMO
Rapid identification of minerals is the key point for enhancing the efficiency of mineral exploration by remote sensing, mineral mapping by remote sensing and many geological investigations. Because of the limitation of technology and other aspects, the amount of models and software concerning rapid identification of minerals is very small. Since 1990s the development in spectrometers and computers has made it possible to apply near infrared spectrum technology to identify minerals. Two models have emerged. Model I is based on analyzing the position of absorption bands, while Model II is founded on waveform matching. In the present paper, characteristic spectrum linear inversion modeling was built. Validated by the data gained from end-members of USGS mineral spectrum library by mixing randomly, this model with the accuracy being approximately 100% is much better than Model I and II. Used to analyze the 23 samples selected in Baogutu area in Xinjiang, the model we built with the accuracy of 64.6% is superior to Model I (the accuracy is 33.8%) and Model II (the accuracy is 8.1%). Though the accuracy of our model is not as high as that of identification by microscope at present, using our model is much more effective and convenient, and there also will be less artificial error and smaller workload. The good performance of our model in the mineral exploration work by remote sensing in Baogutu area in Xinjiang shows wide popularizing prospects.
RESUMO
BACKGROUND: Thyroid carcinoma constitutes the vast majority of all thyroid cancer, most of which is the solid nodule type. No previous studies have examined combining both conventional and elastic sonography to evaluate the diagnostic performance of partially cystic thyroid cancer (PCTC). This retrospective study was designed to evaluate differentiation of PCTC from benign partially cystic nodules with a machine learning-assisted system based on ultrasound (US) and elastography. METHODS: Patients with suspicious partially cystic nodules and finally confirmed were included in the study. We performed conventional US and real-time elastography (RTE). The US features of nodules were recorded. The data set was entered into 6 machine-learning algorithms. Sensitivity, specificity, accuracy, and area under the curve (AUC) were calculated. RESULTS: A total of 177 nodules were included in this study. Among these nodules, 81 were malignant and 96 were benign. Wreath-shaped feature, micro-calcification, and strain ratio (SR) value were the most important imaging features in differential diagnosis. The random forest classifier was the best diagnostic model. CONCLUSIONS: US features of PCTC exhibited unique characteristics. Wreath-shaped partially cystic nodules, especially with the appearance of micro-calcifications and larger SR value, are more likely to be malignant. The random forest classifier might be useful to diagnose PCTC.
RESUMO
To recognize ground objects with infrared spectrum, high frequency noise removing is one of the most important phases in spectrum feature analysis and extraction. A new method for infrared spectrum preprocessing was given combining spectrum continuum processing and Fast Fourier Transform (CFFT). Continuum was firstly removed from the noise polluted infrared spectrum to standardize hyper-spectra. Then the spectrum was transformed into frequency domain (FD) with fast Fourier transform (FFT), separating noise information from target information After noise eliminating from useful information with a low-pass filter, the filtered FD spectrum was transformed into time domain (TD) with fast Fourier inverse transform. Finally the continuum was recovered to the spectrum, and the filtered infrared spectrum was achieved. Experiment was performed for chlorite spectrum in USGS polluted with two kinds of simulated white noise to validate the filtering ability of CFFT by contrast with cubic function of five point (CFFP) in time domain and traditional FFT in frequency domain. A circle of CFFP has limited filtering effect, so it should work much with more circles and consume more time to achieve better filtering result. As for conventional FFT, Gibbs phenomenon has great effect on preprocessing result at edge bands because of special character of rock or mineral spectra, while works well at middle bands. Mean squared error of CFFT is 0. 000 012 336 with cut-off frequency of 150, while that of FFT and CFFP is 0. 000 061 074 with cut-off frequency of 150 and 0.000 022 963 with 150 working circles respectively. Besides the filtering result of CFFT can be improved by adjusting the filter cut-off frequency, and has little effect on working time. The CFFT method overcomes the Gibbs problem of FFT in spectrum filtering, and can be more convenient, dependable, and effective than traditional TD filter methods.
RESUMO
A spectral mixture analysis experiment was designed to compare the spectral unmixing effects of linear spectral mixture analysis (LSMA) and constraint linear spectral mixture analysis (CLSMA). In the experiment, red, green, blue and yellow colors were printed on a coarse album as four end members. Thirty nine mixed samples were made according to each end member's different percent in one pixel. Then, field spectrometer was located on the top of the mixed samples' center to measure spectrum one by one. Inversion percent of each end member in the pixel was extracted using LSMA and CLSMA models. Finally, normalized mean squared error was calculated between inversion and real percent to compare the two models' effects on spectral unmixing. Results from experiment showed that the total error of LSMA was 0.30087 and that of CLSMA was 0.37552 when using all bands in the spectrum. Therefore, LSMA was 0.075 less than that of CLSMA when the whole bands of four end members' spectra were used. On the other hand, the total error of LSMA was 0.28095 and that of CLSMA was 0.29805 after band selection. So, LSMA was 0.017 less than that of CLSMA when bands selection was performed. Therefore, whether all or selected bands were used, the accuracy of LSMA was better than that of CLSMA because during the process of spectrum measurement, errors caused by instrument or human were introduced into the model, leading to that the measured data could not mean the strict requirement of CLSMA and therefore reduced its accuracy: Furthermore, the total error of LSMA using selected bands was 0.02 less than that using the whole bands. The total error of CLSMA using selected bands was 0.077 less than that using the whole bands. So, in the same model, spectral unmixing using selected bands to reduce the correlation of end members' spectra was superior to that using the whole bands.
RESUMO
In order to remove the sawtoothed noise in the spectrum of hyperspectral remote sensing and improve the accuracy of information extraction using spectrum in the present research, the spectrum of vegetation in the USGS (United States Geological Survey) spectrum library was used to simulate the performance of wavelet denoising. These spectra were measured by a custom-modified and computer-controlled Beckman spectrometer at the USGS Denver Spectroscopy Lab. The wavelength accuracy is about 5 nm in the NIR and 2 nm in the visible. In the experiment, noise with signal to noise ratio (SNR) 30 was first added to the spectrum, and then removed by the wavelet denoising approach. For the purpose of finding the optimal parameters combinations, the SNR, mean squared error (MSE), spectral angle (SA) and integrated evaluation coefficient eta were used to evaluate the approach's denoising effects. Denoising effect is directly proportional to SNR, and inversely proportional to MSE, SA and the integrated evaluation coefficient eta. Denoising results show that the sawtoothed noise in noisy spectrum was basically eliminated, and the denoised spectrum basically coincides with the original spectrum, maintaining a good spectral characteristic of the curve. Evaluation results show that the optimal denoising can be achieved by firstly decomposing the noisy spectrum into 3-7 levels using db12, db10, sym9 and sym6 wavelets, then processing the wavelet transform coefficients by soft-threshold functions, and finally estimating the thresholds by heursure threshold selection rule and rescaling using a single estimation of level noise based on first-level coefficients. However, this approach depends on the noise level, which means that for different noise level the optimal parameters combination is also diverse.
RESUMO
In order to highlight target in multispectral remote sensing and overcome the human error caused by threshold, a new method is proposed here. Image of target similarity is firstly calculated by spectral energy level matching (SEM) algorithm and as a band added to original image; Then, band normalization is performed on the new image to reduce the effects caused by the order of magnitude in different bands; Finally, a false color image that highlights the target is made by RGB composed of the first three bands (3, 2, 1) in MNF transformation. Results from the experiment of highlighting the main rock-type tuffaceous siltstone in Hatu area, Xinjiang province, China show that (1) the new method can highlight target for the increment of target's information and weights during the process of transformation by adding a band representing target's similarity to the original image. Therefore, it overcomes the shortcomings existing in the common transformations on space information-although different objects corresponding to special information space are distinguished, targets the authors wanted can not be highlighted yet; (2) The new method can distinguish more objects than original maximum noise fraction (MNF) transformation because it unifies the tone for the same object's type by suppressing none target information using SEM method; (3) In addition to highlighting tuffaceous siltstone in the study area, the new method can be used widely in other fields such as soil, concrete, altered mineral etc.
RESUMO
Based on the principle of mineral generation, structures could provide not only passage ways for ore-forming fluid, but also space for them to aggregate. So, it was very important to study the feature of structures in study area before mineral exploration. In order to highlight structures using multispectral remote sensing data, an algorithm integrating principle component analysis (PCA), maximum noise fraction transformation (MNF) and original image data was proposed here. In the algorithm, the original image was firstly transformed by PCA and MNF; then all bands were normalized to reduce errors caused by different band dimensions, and three bands containing detailed structure information were selected to form the false color image in which structures in study area were highlighted. Results of transformation on enhanced thematic mapper (ETM) data acquired on June 27th 2000 in Hatu area, Xinjiang province, China showed that (1) the transformed image was not only more colorful than the original data, but also more gradational than the original data. (2) The color difference among objects was enhanced by the algorithm. (3) Structrues were highlighted by the algorithm. Therefore, the algorithm's effect of highlighting structures in study area was noticeable.