Self-supervised learning for CT image denoising and reconstruction: a review.

This article reviews the self-supervised learning methods for CT image denoising and reconstruction. Currently, deep learning has become a dominant tool in medical imaging as well as computer vision. In particular, self-supervised learning approaches have attracted great attention as a technique for learning CT images without clean/noisy references. After briefly reviewing the fundamentals of CT image denoising and reconstruction, we examine the progress of deep learning in CT image denoising and reconstruction. Finally, we focus on the theoretical and methodological evolution of self-supervised learning for image denoising and reconstruction.

Development of a corn flour certified reference material for the accurate determination of zearalenone.

A certified reference material (CRM, KRISS 108-01-002) for zearalenone in corn flour was developed to assure reliable and accurate measurements in testing laboratories. Commercially available corn flour underwent freeze-drying, pulverization, sieving, and homogenization. The final product was packed in amber bottles, approximately 14 g per unit, and preserved at -70 °C. 13C18-Zearalenone was used as an internal standard (IS) for the certification of zearalenone by isotope-dilution liquid chromatography-tandem mass spectrometry (ID-LC‒MS/MS) and for the analysis of α-zearalenol, ß-zearalenol, and zearalanone by LC‒MS/MS. The prepared CRM was sufficiently homogeneous, as the among-unit relative standard deviation for each mycotoxin ranged from 2.2 to 5.7 %. Additionally, the stability of the mycotoxins in the CRM was evaluated under different temperature conditions and scheduled test periods, including storage at -70°C, -20°C, and 4°C and room temperature for up to 12 months, 6 months, and 1 month, respectively. The content of each target mycotoxin in the CRM remained stable throughout the monitoring period at each temperature. Zearalenone content (153.6 ± 8.0 µg/kg) was assigned as the certified value. Meanwhile, the contents of α-zearalenol (1.30 ± 0.17 µg/kg), ß-zearalenol (4.75 ± 0.33 µg/kg), and zearalanone (2.09 ± 0.16 µg/kg) were provided as informative values.

Prognostic Performance of Sequential Organ Failure Assessment, Acute Physiology and Chronic Health Evaluation III, and Simplified Acute Physiology Score II Scores in Patients with Suspected Infection According to Intensive Care Unit Type.

We investigated the prognostic performance of scoring systems by the intensive care unit (ICU) type. This was a retrospective observational study using data from the Marketplace for Medical Information in the Intensive Care IV database. The primary outcome was in-hospital mortality. We obtained Sequential Organ Failure Assessment (SOFA), Acute Physiology and Chronic Health Evaluation (APACHE) III, and Simplified Acute Physiology Score (SAPS) II scores in each ICU type. Prognostic performance was evaluated with the area under the receiver operating characteristic curve (AUROC) and was compared among ICU types. A total of 29,618 patients were analyzed, and the in-hospital mortality was 12.4%. The overall prognostic performance of APACHE III was significantly higher than those of SOFA and SAPS II (0.807, [95% confidence interval, 0.799-0.814], 0.785 [0.773-0.797], and 0.795 [0.787-0.811], respectively). The prognostic performance of SOFA, APACHE III, and SAPS II scores was significantly different between ICU types. The AUROC ranges of SOFA, APACHE III, and SAPS II were 0.723-0.826, 0.728-0.860, and 0.759-0.819, respectively. The neurosurgical and surgical ICUs had lower prognostic performance than other ICU types. The prognostic performance of scoring systems in patients with suspected infection is significantly different according to ICU type. APACHE III systems have the highest prediction performance. ICU type may be a significant factor in the prognostication.

Self-supervised denoising of projection data for low-dose cone-beam CT.

BACKGROUND: Convolutional neural networks (CNNs) have shown promising results in image denoising tasks. While most existing CNN-based methods depend on supervised learning by directly mapping noisy inputs to clean targets, high-quality references are often unavailable for interventional radiology such as cone-beam computed tomography (CBCT). PURPOSE: In this paper, we propose a novel self-supervised learning method that reduces noise in projections acquired by ordinary CBCT scans. METHODS: With a network that partially blinds input, we are able to train the denoising model by mapping the partially blinded projections to the original projections. Additionally, we incorporate noise-to-noise learning into the self-supervised learning by mapping the adjacent projections to the original projections. With standard image reconstruction methods such as FDK-type algorithms, we can reconstruct high-quality CBCT images from the projections denoised by our projection-domain denoising method. RESULTS: In the head phantom study, we measure peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) values of the proposed method along with the other denoising methods and uncorrected low-dose CBCT data for a quantitative comparison both in projection and image domains. The PSNR and SSIM values of our self-supervised denoising approach are 27.08 and 0.839, whereas those of uncorrected CBCT images are 15.68 and 0.103, respectively. In the retrospective study, we assess the quality of interventional patient CBCT images to evaluate the projection-domain and image-domain denoising methods. Both qualitative and quantitative results indicate that our approach can effectively produce high-quality CBCT images with low-dose projections in the absence of duplicate clean or noisy references. CONCLUSIONS: Our self-supervised learning strategy is capable of restoring anatomical information while efficiently removing noise in CBCT projection data.

Occurrence of zearalenone and its major metabolites in cereal flour from Korea.

The occurrence of zearalenone (ZEN) and its metabolites (α-zearalenol (α-ZEL), ß-zearalenol (ß-ZEL), α-zearalanol (α-ZAL), ß-zearalanol (ß-ZAL), and zearalanone (ZAN)) was investigated in 78 cereal flour from Korea using UHPLC-MS/MS. Among these mycotoxins, ZEN was the most abundant in the analyzed samples at an incidence rate of 41% and concentration range of 0.5-536 µg/kg. The highest contamination and incidence rate of ZEN were found in corn flour samples, while oat flour samples showed the lowest contamination and incidence rate of this mycotoxin. α-ZEL, ß-ZEL, and ZAN were detected only in corn flour samples but at lower frequencies of 23%, 17%, and 15%, respectively, while α-ZAL and ß-ZAL were not detected in any sample. To the best of our knowledge, this is the first investigation of the simultaneous occurrence of ZEN and its major metabolites in commercially available cereal flour from Korea. Among the tested samples, only four were contaminated with ZEN at levels exceeding the maximum regulatory level established in Korea. The co-occurrence of ZEN, α-ZEL, ß-ZEL, and ZAN was observed in 14% of all samples. Although ZEN metabolites were detected at relatively lower levels than ZEN, the relatively high co-occurrence rate of those mycotoxins is of significant concern from a food safety perspective, since they can synergistically contribute to the overall toxicity and estrogenic effects.

Type B Trichothecenes in Cereal Grains and Their Products: Recent Advances on Occurrence, Toxicology, Analysis and Post-Harvest Decontamination Strategies.

Type B trichothecenes (deoxynivalenol, nivalenol, 3-acetyldeoxynivalenol, 15-acetyldeoxynivalenol) and deoxynivalenol-3-glucoside (DON-3G) are secondary toxic metabolites produced mainly by mycotoxigenic Fusarium fungi and have been recognized as natural contaminants in cereals and cereal-based foods. The latest studies have proven the various negative effects of type B trichothecenes on human health. Due to the widespread occurrence of Fusarium species, contamination by these mycotoxins has become an important aspect for public health and agro-food systems worldwide. Hence, their monitoring and surveillance in various foods have received a significant deal of attention in recent years. In this review, an up-to-date overview of the occurrence profile of major type B trichothecenes and DON-3G in cereal grains and their toxicological implications are outlined. Furthermore, current trends in analytical methodologies for their determination are overviewed. This review also covers the factors affecting the production of these mycotoxins, as well as the management strategies currently employed to mitigate their contamination in foods. Information presented in this review provides good insight into the progress that has been achieved in the last years for monitoring type B trichothecenes and DON-3G, and also would help the researchers in their further investigations on metabolic pathway analysis and toxicological studies of these Fusarium mycotoxins.

Accurate determination of four tetracycline residues in chicken meat by isotope dilution-liquid chromatography/tandem mass spectrometry.

An analytical method based on isotope dilution-liquid chromatography-tandem mass spectrometry (ID-LC‒MS/MS) was developed to accurately determine four representative tetracyclines (tetracycline, chlortetracycline, doxycycline, and oxytetracycline) in chicken meat. Tetracyclines are known to have a great tendency for epimerization and keto-enol tautomerism, which often provoke major challenges in their determination. Since this isomerization was found to be unavoidable during the whole chain of the current analysis, the total content (µg kg‒1) of individual tetracycline was quantified as a sum of each parent compound and its respective isomeric forms. Using this approach in combination with IDMS analysis, more consistent, accurate, and reproducible measurement results for the four tetracyclines in chicken meat were acquired. LC-MS/MS conditions and sample preparation processes were comprehensively optimized to minimize the chelating effect of tetracyclines and possible co-extracted interferences. Details of the sample preparation scheme, LC‒MS/MS detection, calculation equation, and method validation are described in this article. The method provided very good accuracy (97.7-102.6%) for all analytes across the concentration range of 10-200 µg kg‒1, with relative standard deviations for intra-day and inter-day precision of less than 4%. The limits of quantification were below 0.2 µg kg‒1, demonstrating the high sensitivity of the method. Furthermore, the measurement uncertainty was generally below 5.5%. Hence, the established method exhibits high-order metrological quality with superior performance over various existing methodologies. Moreover, this method can provide references for general food testing laboratories close to and far below the established maximum residue limits (100 µg kg‒1) for animal muscle tissues.

Development of a certified reference material for the accurate determination of type B trichothecenes in corn.

A corn flour certified reference material (KRISS CRM 108-01-011) was developed to ensure accurate and reliable measurements of type B trichothecenes (deoxynivalenol, nivalenol, 3-acetyldeoxynivalenol, 15-acetyldeoxynivalenol), and deoxynivalenol-3-glucoside. The material was freeze-dried, ground, sieved, and well-mixed. The final produced CRM was packaged at 14 g per unit and stored at -70 °C. The certification was performed using isotope dilution-liquid chromatography-tandem mass spectrometry (ID-LC-MS/MS). For simultaneous characterization and homogeneity assessments, ten units were randomly selected and analyzed. The among-unit relative standard deviations were below 1 % for all mycotoxins, indicating excellent homogeneity of the CRM. The stability of the CRM was also assessed at various designated temperatures and test periods. The uncertainties of the certified values varied between 2.4 % and 6.2 %, thereby confirming their higher-order metrological quality to provide references for testing laboratories. In case of deoxynivalenol-3-glucoside, an information value was assigned due to the lack of its traceability to the SI units.

A Bounding-Box Regression Model for Colorectal Tumor Detection in CT Images Via Two Contrary Networks.

The field of medical image analysis has been attracted to deep learning. Various deep learning-based techniques have been introduced to aid diagnosis in the CT image of the patient. The auxiliary model for diagnosis that we proposed is to detect colorectal tumors in the CT image. The model is combined with two contrary networks of 'Detection Transformer" and 'Hourglass". Furthermore., to improve the performance of the model., we propose an efficient connection method for two contrary models by using intermediate prediction information. A total of 3.,509 patients (193.,567 CT images) were applied to the experiment and our model outperforms the conventional models in colorectal tumor detection. Clinical Relevance - The proposed model in this paper automatically detects colorectal tumors and provides the bounding box in the CT images. Colorectal tumor is one of the common diseases. In addition, the mortality rate is so high that in-time treatment is required. The model we present here has a sensitivity (or recall) of 84.73 % for tumor detection and a precision of 88.25 % in the patient CT data. The in-slice performance of the tumor detection shows an IoU of 0.56, a sensitivity of 0.67, and a precision of 0.68.

The DEXA-SEPSIS study protocol: a phase II randomized double-blinded controlled trial of the effect of early dexamethasone in high-risk sepsis patients.

OBJECTIVE: Steroids are used in cases of sepsis, especially in patients experiencing septic shock. However, clinical trials to date have reported contradictory results. Different patient endotypes and variations in the type and dose of steroid may be at fault for this discrepancy, and further investigation is warranted. In this paper, we propose a new DEXA-SEPSIS study design. METHODS: We plan to conduct a multicenter, double-blinded randomized pilot study (DEXA-SEPSIS) investigating the feasibility and safety of early use of dexamethasone in sepsis. Participants will be high-risk septic patients presenting to the emergency department with a systolic blood pressure of <90 mmHg or serum lactate level of >2 mmol/L. Participants will be randomized to the following three groups: control, 0.1 mg/kg of dexamethasone, or 0.2 mg/kg of dexamethasone per day for 1 to 2 days. The primary outcome will be 28-day mortality. Secondary outcomes will include time to septic shock, shock reversal, additional steroid administration, number of ventilator-free days, use of continuous renal-replacement therapy, length of stay in the intensive care unit and/or hospital, delta Sequential Organ Failure Assessment score on days 3 and 7, superinfection, gastrointestinal bleeding, hypernatremia, and hyperglycemia. DISCUSSION: The DEXA-SEPSIS study will provide insight regarding the feasibility and safety of early use of dexamethasone in high-risk sepsis. The results could provide data to design a future phase III study. TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT05136560.

A Comparative Study between Image- and Projection-Domain Self-Supervised Learning for Ultra Low-Dose CBCT.

We compare image domain and projection domain denoising approaches with self-supervised learning for ultra low-dose cone-beam CT (CBCT), where number of detected x-ray photons is significantly low. For image-domain self-supervised denoising, we first reconstruct CBCT images with the standard filtered backprojection. For model training, we use blind-spot filtering to partially blind images and recover the blind spots. For projection-domain self-supervised denoising, we regard the post-log projections as training examples of convolutional neural network. From experimental results with various low-dose CBCT settings, the projection-domain denoiser outperforms the image-domain denoiser both in image quality and accuracy for ultra low-dose CBCT.

Development of isotope dilution-liquid chromatography/tandem mass spectrometry for the accurate determination of zearalenone and its metabolites in corn.

A method using isotope dilution-liquid chromatography-tandem mass spectrometry (ID-LC/MS/MS) was developed for the accurate determination of zearalenone (ZEN) and its five major metabolites in corn. 13C- or 2H-labeled analogues of the target mycotoxins were used as internal standards. As the immunoaffinity columns used demonstrated selectivity to a specific chiral isomer of a racemic mixture of zearalanone-d6, a clean-up cartridge without stereoselectivity (Mycosep 226 column) was selected for the same recovery of the analyte and its internal standard with adequate elimination of matrix interferences. The method demonstrated sufficient selectivity, sensitivity, accuracy and precision over a concentration range of 20-400 µg/kg. The limit of detections and limit of quantifications were 0.14-0.33 µg/kg and 0.45-1.11 µg/kg, respectively. The accuracy values were 96.7%-103.6%, with intra and inter-day precisions of less than 3% and 4%, respectively. The expanded measurement uncertainty was less than 7% (with a 95% confidence level).

Development of isotope dilution-liquid chromatography/tandem mass spectrometry as a candidate reference method for the accurate determination of patulin in apple products.

The incidence of patulin (PAT) in fruit products is a worldwide concern due to its acute and chronic toxic effects. Therefore, accurate and reliable PAT measurements are important for preventing consumer health risks. Our previous method, which was based on a common technique that uses ethyl acetate extraction and liquid chromatography-tandem mass spectrometry with isotope dilution (ID-LC-MS/MS), has shown great performance for the determination of PAT in apple products. However, prolonged extraction times and multistep clean-up processes were required to sufficiently eliminate the matrix interferences. Herein, a feasible alternative ID-LC-MS/MS method was successfully established, employing simplified and reliable sample preparation steps. The clean-up process was performed using molecularly imprinted polymer-solid-phase extraction (MIP-SPE) cartridges, which eliminated matrix interferences and facilitated the trace quantification. While the previous method used a multimode LC column for the retention of polar patulin, the current method used a UPLC HSS T3 column, which further improved the peak sharpness and reduced the run time. The method was validated by measuring fortified samples in the concentration range of 5‒100 µg/kg. The accuracy varied between 97.8 and 102.0%, with relative standard deviation for interday and intraday precision being below 3%. The measurement uncertainty was lower than 4% (at a 95% level of confidence). Therefore, this method demonstrated adequate metrological quality with greatly enhanced performance over various reported methods. Additional key benefits of this method are easy manipulation, short preparation time, and lower consumption of hazardous solvents. Finally, the method was successfully applied to commercially available apple-based products.

Self-supervised Projection Denoising for Low-Dose Cone-Beam CT.

We consider the problem of denoising low-dose xray projections for cone-beam CT, where x-ray measurements are typically modeled as signal corrupted by Poisson noise. Since each projection view is a 2D image, we regard the lowdose projection views as examples to train a convolutional neural network. For self-supervised training without ground truth, we partially blind noisy projections and train the denoising model to recover the blind spots of projection views. From the projection views denoised by the learned model, we can reconstruct a high-quality 3D volume with a reconstruction algorithm such as the standard filtered backprojection. Through a series of phantom experiments, our self-supervised denoising approach simultaneously reduces noise level and restores structural information in cone-beam CT images.

Weakly Supervised Attention Map Training for Histological Localization of Colonoscopy Images.

We consider the problem of training a convolutional neural network for histological localization of colorectal lesions from imperfectly annotated datasets. Given that we have a colonoscopic image dataset for 4-class histology classification and another dataset originally dedicated to polyp segmentation, we propose a weakly supervised learning approach to histological localization by training with the two different types of datasets. With the classification dataset, we first train a convolutional neural network to classify colonoscopic images into 4 different histology categories. By interpreting the trained classifier, we can extract an attention map corresponding to the predicted class for each colonoscopy image. We further improve the localization accuracy of attention maps by training the model to focus on lesions under the guidance of the polyp segmentation dataset. The experimental results show that the proposed approach simultaneously improves histology classification and lesion localization accuracy.

Prediction of the histology of colorectal neoplasm in white light colonoscopic images using deep learning algorithms.

The treatment plan of colorectal neoplasm differs based on histology. Although new endoscopic imaging systems have been developed, there are clear diagnostic thresholds and requirements in using them. To overcome these limitations, we trained convolutional neural networks (CNNs) with endoscopic images and developed a computer-aided diagnostic (CAD) system which predicts the pathologic histology of colorectal adenoma. We retrospectively collected colonoscopic images from two tertiary hospitals and labeled 3400 images into one of 4 classes according to the final histology: normal, low-grade dysplasia, high-grade dysplasia, and adenocarcinoma. We implemented a CAD system based on ensemble learning with three CNN models which transfer the knowledge learned from common digital photography images to the colonoscopic image domain. The deep learning models were trained to classify the colorectal adenoma into these 4 classes. We compared the outcomes of the CNN models to those of two endoscopist groups having different years of experience, and visualized the model predictions using Class Activation Mapping. In our multi-center study, our CNN-CAD system identified the histology of colorectal adenoma with as sensitivity 77.25%, specificity of 92.42%, positive predictive value of 77.16%, negative predictive value of 92.58% averaged over the 4 classes, and mean diagnostic time of 0.12 s per image. Our experiments demonstrate that the CNN-CAD showed a similar performance to that of endoscopic experts and outperformed that of trainees. The model visualization results also showed reasonable regions of interest to explain the classification decisions of CAD systems. We suggest that CNN-CAD system can predict the histology of colorectal adenoma.

Development of isotope dilution-liquid chromatography/tandem mass spectrometry for the accurate determination of type-A trichothecenes in grains.

We report the development of a highly accurate method based on isotope dilution liquid chromatography-tandem mass spectrometry (ID-LC-MS/MS) for the simultaneous determination of four major type-A trichothecenes in cereal grains. Uniformly labeled [13C] analogues of the target analytes were used as internal standards. An expedient sample preparation procedure was established. After extraction with acetonitrile/water (84:16; v/v), further clean-up was performed using MycoSep 227 solid-phase extraction cartridges. Unlike the commonly used immunoaffinity columns having strong selectivity for only certain target analytes, the cartridges allowed the simultaneous recovery of all four mycotoxins and efficient elimination of co-extracted matrix interferences. The ID-LC-MS/MS method exhibited very good analytical performance in the concentration range of 10-200 µg/kg; accuracy ranged from 97 to 103% with intra-day and inter-day relative standard deviations of less than 5% and 4%, respectively. Measurement uncertainties were generally below 5%. The applicability of the method was assessed by measuring the target mycotoxins in several samples at sub-µg/kg levels.

Computer-Aided Diagonosis for Colorectal Cancer using Deep Learning with Visual Explanations.

Detection, diagnosis, and removal of colorectal neoplasms are well-accepted colorectal cancer prevention methods. Although promising endoscopic imaging techniques including narrow-band imaging have been developed, these techniques are operator-dependent and interpretations of the results may vary. To overcome these limitations, we applied deep learning to develop a computer-aided diagnostic (CAD) system of colorectal adenoma. We collected and divided 3000 colonoscopic images into 4 categories according to the final pathology, normal, low-grade dysplasia, high-grade dysplasia, and adenocarcinoma. We implemented three convolutional neural networks (CNNs) using Inception-v3, ResNet-50, and DenseNet-161 as baseline models. We further altered the models using several strategies: replacement of the top layer, transfer learning from pre-trained models, fine-tuning of the model weights, rebalancing and augmentation of the training data, and 10-fold cross-validation. We compared the outcomes of the three CNN models to those of two endoscopist groups having different years of experience, and visualized the model predictions using Class Activation Mapping (CAM). The CNN-CAD achieved the best performance in our experiments with a 92.48% classification accuracy rate. The CNN-CAD results showed a better performance in all criteria than those of endoscopic experts. The model visualization results showed reasonable regions of interest to explain pathology classification decisions. We demonstrated that CNN-CAD can distinguish the pathology of colorectal adenoma, yielding better outcomes than the endoscopic experts group.

Statistical Image Restoration for Low-Dose CT using Convolutional Neural Networks.

Deep learning has recently attracted widespread interest as a means of reducing noise in low-dose CT (LDCT) images. Deep convolutional neural networks (CNNs) are typically trained to transfer high-quality image features of normal-dose CT (NDCT) images to LDCT images. However, existing deep learning approaches for denoising LDCT images often overlook the statistical property of CT images. In this paper, we propose an approach to statistical image restoration for LDCT using deep learning (StatCNN). We introduce a loss function to incorporate the noise property in the image domain derived from the noise statistics in the sinogram domain. In order to capture the spatially-varying statistics of axial CT images, we increase the receptive fields of the proposed network to cover full-size CT slices. In addition, the proposed network utilizes z-directional correlation by taking multiple consecutive CT slices as input. For performance evaluation, the proposed network was thoroughly trained and tested by leave-one-out cross-validation with a dataset consisting of LDCT-NDCT image pairs. The experimental results showed that the denoising networks successfully reduced the noise level and restored the image details without adding artifacts. This study demonstrates that the statistical deep learning approach can transfer the image style from NDCT images to LDCT images without loss of anatomical information.

Development of an infant formula certified reference material for the analysis of organic nutrients.

Infant formula certified reference material (CRM, KRISS CRM 108-02-003) were developed for the analysis of organic nutrients. The CRM is a milk-based infant formula powder, packaged at 14 g per unit. Ten thousand units were prepared and stored at -70 °C. For the certification of each nutrient, ten units were analyzed for simultaneous value-assignment and homogeneity test. Analytical methods used were isotope dilution mass spectrometry (IDMS) based on liquid chromatography mass spectrometer (LC/MS) or gas chromatography mass spectrometer (GC/MS) as higher-order reference methods.13 vitamins, 3 fatty acids, and total cholesterol were certified. The between-unit relative standard deviation of measurement results for each nutrient ranged 0.2% to 2.5%, showing very good homogeneity. The expanded relative uncertainties of the certified values ranged from 1% to 8%, indicating that they have higher-order metrological quality. The values of proximates (proteins, lipids, carbohydrates, water, and ash) were assigned through inter-laboratory comparisons.