Clinical Utility of the Tokyo Guidelines 2018 for Acute Cholangitis in the Emergency Department and Comparison with Novel Markers (Neutrophil-to-Lymphocyte and Blood Nitrogen Urea-to-Albumin Ratios).

Introduction: The Tokyo Guidelines 2018 (TG2018) is a scoring system used to recommend the clinical management of AC. However, such a scoring system must incorporate a variety of clinical outcomes of acute cholangitis (AC). In an emergency department (ED)-based setting, where efficiency and practicality are highly desired, clinicians may find the application of various parameters challenging. The neutrophil-to-lymphocyte ratio (NLR) and blood urea nitrogen-to-albumin ratio (BAR) are relatively common biomarkers used to assess disease severity. This study evaluated the potential value of TG2018 scores measured in an ED to predict a variety of clinical outcomes. Furthermore, the study also compared TG2018 scores with NLR and BAR scores to demonstrate their usefulness. Methods: This retrospective observational study was performed in an ED. In total, 502 patients with AC visited the ED between January 2016 and December 2021. The primary endpoint was to evaluate whether the TG2018 scoring system measured in the ED was a predictor of intensive care, long-term hospital stays (≥14 days), percutaneous transhepatic biliary drainage (PTBD) during admission care, and endotracheal intubation (ETI). Results: The analysis included 81 patients requiring intensive care, 111 requiring long-term hospital stays (≥14 days), 49 requiring PTBD during hospitalization, and 14 requiring ETI during hospitalization. For the TG2018 score, the adjusted OR (aOR) using (1) as a reference was 23.169 (95% CI: 9.788-54.844) for (3) compared to (1). The AUC of the TG2018 for the need for intensive care was 0.850 (95% CI: 0.815-0.881) with a cutoff of >2. The AUC for long-term hospital stays did not exceed 0.7 for any of the markers. the AUC for PTBD also did not exceed 0.7 for any of the markers. The AUC for ETI was the highest for BAR at 0.870 (95% CI: 0.837-0.899) with a cutoff value of >5.2. Conclusions: The TG2018 score measured in the ED helps predict various clinical outcomes of AC. Other novel markers such as BAR and NLR are also associated, but their explanatory power is weak.

Evaluation of Fused Deposition Modeling Materials for 3D-Printed Container of Dosimetric Polymer Gel.

Accurate dosimetric verification is becoming increasingly important in radiotherapy. Although polymer gel dosimetry may be useful for verifying complex 3D dose distributions, it has limitations for clinical application due to its strong reactivity with oxygen and other contaminants. Therefore, it is important that the material of the gel storage container blocks reaction with external contaminants. In this study, we tested the effect of air and the chemical permeability of various polymer-based 3D printing materials that can be used as gel containers. A methacrylic acid, gelatin, and tetrakis (hydroxymethyl) phosphonium chloride gel was used. Five types of printing materials that can be applied to the fused deposition modeling (FDM)-type 3D printer were compared: acrylonitrile butadiene styrene (ABS), co-polyester (CPE), polycarbonate (PC), polylactic acid (PLA), and polypropylene (PP) (reference: glass vial). The map of R2 (1/T2) relaxation rates for each material, obtained from magnetic resonance imaging scans, was analyzed. Additionally, response histograms and dose calibration curves from the R2 map were evaluated. The R2 distribution showed that CPE had sharper boundaries than the other materials, and the profile gradient of CPE was also closest to the reference vial. Histograms and dose calibration showed that CPE provided the most homogeneous and the highest relative response of 83.5%, with 8.6% root mean square error, compared with the reference vial. These results indicate that CPE is a reasonable material for the FDM-type 3D printing gel container.

A multimodal screening system for elderly neurological diseases based on deep learning.

In this paper, we propose a deep-learning-based algorithm for screening neurological diseases. We proposed various examination protocols for screening neurological diseases and collected data by video-recording persons performing these protocols. We converted video data into human landmarks that capture action information with a much smaller data dimension. We also used voice data which are also effective indicators of neurological disorders. We designed a subnetwork for each protocol to extract features from landmarks or voice and a feature aggregator that combines all the information extracted from the protocols to make a final decision. Multitask learning was applied to screen two neurological diseases. To capture meaningful information about these human landmarks and voices, we applied various pre-trained models to extract preliminary features. The spatiotemporal characteristics of landmarks are extracted using a pre-trained graph neural network, and voice features are extracted using a pre-trained time-delay neural network. These extracted high-level features are then passed onto the subnetworks and an additional feature aggregator that are simultaneously trained. We also used various data augmentation techniques to overcome the shortage of data. Using a frame-length staticizer that considers the characteristics of the data, we can capture momentary tremors without wasting information. Finally, we examine the effectiveness of different protocols and different modalities (different body parts and voice) through extensive experiments. The proposed method achieves AUC scores of 0.802 for stroke and 0.780 for Parkinson's disease, which is effective for a screening system.

The D-Dimer to Troponin Ratio Is a Novel Marker for the Differential Diagnosis of Thoracic Acute Aortic Syndrome from Non-ST Elevation Myocardial Infarction.

BACKGROUND: Thoracic acute aortic syndrome (AAS) and non-ST elevation myocardial infarction (NSTEMI) have similar clinical presentations, making them difficult to differentiate. This study aimed to identify useful biomarkers for the differential diagnosis of thoracic AAS and NSTEMI. METHODS: This was a retrospective observational study. PARTICIPANTS: consecutive adult patients who visited the emergency department for acute chest pain between January 2015 and December 2021 diagnosed with thoracic AAS or NSTEMI. Clinical variables, including D-dimer (µg/mL) and high-sensitivity troponin T (ng/mL, hs-TnT) levels, were compared between the groups. RESULTS: A total of 52 (30.1%) and 121 (69.9%) patients were enrolled in the thoracic AAS and NSTEMI groups, respectively. Logistic regression analysis revealed that the D-dimer to hs-TnT (D/T) ratio (odds ratio (OR), 1.038; 95% confidence interval (CI), 1.020-1.056; p < 0.001) and the thrombolysis in myocardial infarction (TIMI) score (OR, 0.184; 95% CI, 0.054-0.621; p = 0.006) were associated with thoracic AAS. The D/T ratio had an area under the receiver operating characteristic curve (AUC) of 0.973 (95% CI, 0.930-0.998), and the optimal cutoff value was 81.3 with 91.4% sensitivity and 96.2% specificity. The TIMI score had an AUC of 0.769 (95% CI, 0.644-0.812), and the optimal cutoff value was 1.5 with 96.7% sensitivity and 38.5% specificity. CONCLUSION: the D/T ratio may be a simple and useful parameter for differentiating thoracic AAS from NSTEMI.

Ornithine aminotransferase supports polyamine synthesis in pancreatic cancer.

There is a need to develop effective therapies for pancreatic ductal adenocarcinoma (PDA), a highly lethal malignancy with increasing incidence1 and poor prognosis2. Although targeting tumour metabolism has been the focus of intense investigation for more than a decade, tumour metabolic plasticity and high risk of toxicity have limited this anticancer strategy3,4. Here we use genetic and pharmacological approaches in human and mouse in vitro and in vivo models to show that PDA has a distinct dependence on de novo ornithine synthesis from glutamine. We find that this process, which is mediated through ornithine aminotransferase (OAT), supports polyamine synthesis and is required for tumour growth. This directional OAT activity is usually largely restricted to infancy and contrasts with the reliance of most adult normal tissues and other cancer types on arginine-derived ornithine for polyamine synthesis5,6. This dependency associates with arginine depletion in the PDA tumour microenvironment and is driven by mutant KRAS. Activated KRAS induces the expression of OAT and polyamine synthesis enzymes, leading to alterations in the transcriptome and open chromatin landscape in PDA tumour cells. The distinct dependence of PDA, but not normal tissue, on OAT-mediated de novo ornithine synthesis provides an attractive therapeutic window for treating patients with pancreatic cancer with minimal toxicity.

Deep-Learning-Based Stroke Screening Using Skeleton Data from Neurological Examination Videos.

According to the Korea Institute for Health and Social Affairs, in 2017, the elderly, aged 65 or older, had an average of 2.7 chronic diseases per person. The concern for the medical welfare of the elderly is increasing due to a low birth rate, an aging population, and the lack of medical personnel. The demand for services that take user age, cognitive capacity, and difficulty into account is rising. As a result, there is an increased demand for smart healthcare systems that can lower hospital admissions and offer patients individualized care. This has motivated us to develop an AI system that can easily screen and manage neurological diseases through videos. As neurological diseases can be diagnosed by visual analysis to some extent, in this study, we set out to estimate the possibility of a person having a neurological disease from videos. Among neurological diseases, we focus on stroke because it is a common condition in the elderly population and results in high mortality and morbidity worldwide. The proposed method consists of three steps: (1) transforming neurological examination videos into landmark data, (2) converting the landmark data into recurrence plots, and (3) estimating the possibility of a stroke using deep neural networks. Major features, such as the hand, face, pupil, and body movements of a person are extracted from test videos taken under several neurological examination protocols using deep-learning-based landmark extractors. Sequences of these landmark data are then converted into recurrence plots, which can be interpreted as images. These images can be fed into convolutional neural networks to classify stroke using feature-fusion techniques. A case study of the application of a disease screening test to assess the capability of the proposed method is presented.

Point Cloud Resampling by Simulating Electric Charges on Metallic Surfaces.

3D point cloud resampling based on computational geometry is still a challenging problem. In this paper, we propose a point cloud resampling algorithm inspired by the physical characteristics of the repulsion forces between point electrons. The points in the point cloud are considered as electrons that reside on a virtual metallic surface. We iteratively update the positions of the points by simulating the electromagnetic forces between them. Intuitively, the input point cloud becomes evenly distributed by the repulsive forces. We further adopt an acceleration and damping terms in our simulation. This system can be viewed as a momentum method in mathematical optimization and thus increases the convergence stability and uniformity performance. The net force of the repulsion forces may contain a normal directional force with respect to the local surface, which can make the point diverge from the surface. To prevent this, we introduce a simple restriction method that limits the repulsion forces between the points to an approximated local plane. This approach mimics the natural phenomenon in which positive electrons cannot escape from the metallic surface. However, this is still an approximation because the surfaces are often curved rather than being strict planes. Therefore, we project the points to the nearest local surface after the movement. In addition, we approximate the net repulsion force using the K-nearest neighbor to accelerate our algorithm. Furthermore, we propose a new measurement criterion that evaluates the uniformity of the resampled point cloud to compare the proposed algorithm with baselines. In experiments, our algorithm demonstrates superior performance in terms of uniformization, convergence, and run-time.

Adaptation of pancreatic cancer cells to nutrient deprivation is reversible and requires glutamine synthetase stabilization by mTORC1.

Pancreatic ductal adenocarcinoma (PDA) is a lethal, therapy-resistant cancer that thrives in a highly desmoplastic, nutrient-deprived microenvironment. Several studies investigated the effects of depriving PDA of either glucose or glutamine alone. However, the consequences on PDA growth and metabolism of limiting both preferred nutrients have remained largely unknown. Here, we report the selection for clonal human PDA cells that survive and adapt to limiting levels of both glucose and glutamine. We find that adapted clones exhibit increased growth in vitro and enhanced tumor-forming capacity in vivo. Mechanistically, adapted clones share common transcriptional and metabolic programs, including amino acid use for de novo glutamine and nucleotide synthesis. They also display enhanced mTORC1 activity that prevents the proteasomal degradation of glutamine synthetase (GS), the rate-limiting enzyme for glutamine synthesis. This phenotype is notably reversible, with PDA cells acquiring alterations in open chromatin upon adaptation. Silencing of GS suppresses the enhanced growth of adapted cells and mitigates tumor growth. These findings identify nongenetic adaptations to nutrient deprivation in PDA and highlight GS as a dependency that could be targeted therapeutically in pancreatic cancer patients.

Reconstruct as Far as You Can: Consensus of Non-Rigid Reconstruction from Feasible Regions.

Much progress has been made for non-rigid structure from motion (NRSfM) during the last two decades, which made it possible to provide reasonable solutions for synthetically-created benchmark data. In order to utilize these NRSfM techniques in more realistic situations, however, we are now facing two important problems that must be solved: First, general scenes contain complex deformations as well as multiple objects, which violates the usual assumptions of previous NRSfM proposals. Second, there are many unreconstructable regions in the video, either because of the discontinued tracks of 2D trajectories or those regions static towards the camera, which require careful manipulations. In this paper, we show that a consensus-based reconstruction framework can handle these issues effectively. Even though the entire scene is complex, its parts usually have simpler deformations, and even though there are some unreconstructable parts, they can be weeded out to reduce their harmful effect on the entire reconstruction. The main difficulty of this approach lies in identifying appropriate parts, however, it can be effectively avoided by sampling parts stochastically and then aggregate their reconstructions afterwards. Experimental results show that the proposed method renews the state-of-the-art for popular benchmark data under much harsher environments, i.e., narrow camera view ranges, and it can reconstruct video-based real-world data effectively for as many areas as it can without an elaborated user input.

Author Correction: PI3K/AKT activation induces PTEN ubiquitination and destabilization accelerating tumourigenesis.

A Correction to this paper has been published: https://doi.org/10.1038/s41467-020-20178-0.

Clinical implementation of a wide-field electron arc technique with a scatterer for widespread Kaposi's sarcoma in the distal extremities.

A novel wide-field electron arc technique with a scatterer is implemented for widespread Kaposi's sarcoma (KS) in the distal extremities. Monte Carlo beam modeling for electron arc beams was established to achieve <2% deviation from the measurements, and used for dose calculation. MC-based electron arc plan was performed using CT images of a foot and leg mimicking phantom and compared with in-vivo measurement data. We enrolled one patient with recurrent KS on the lower extremities who had been treated with photon radiation therapy. The 4- and 6-MeV electron arc plans were created, and then compared to two photon plans: two opposite photon beam and volumetric modulated arc with bolus. Compared to the two photon techniques, the electron arc plans resulted in superior dose saving to normal organs beneath the skin region, although it shows inferior coverage and homogeneity for PTV. The electron arc treatment technique with scatterer was successfully implemented for the treatment of widespread KS in the distal extremities with lower radiation exposure to the normal organs beyond the skin lesions, which could be a treatment option for recurrent skin cancer in the extremities.

Nonparametric Estimation of Probabilistic Membership for Subspace Clustering.

Recent advances of subspace clustering have provided a new way of constructing affinity matrices for clustering. Unlike the kernel-based subspace clustering, which needs tedious tuning among infinitely many kernel candidates, the self-expressive models derived from linear subspace assumptions in modern subspace clustering methods are rigorously combined with sparse or low-rank optimization theory to yield an affinity matrix as a solution of an optimization problem. Despite this nice theoretical aspect, the affinity matrices of modern subspace clustering have quite different meanings from the traditional ones, and even though the affinity matrices are expected to have a rough block-diagonal structure, it is unclear whether these are good enough to apply spectral clustering. In fact, most of the subspace clustering methods perform some sort of affinity value rearrangement to apply spectral clustering, but its adequacy for the spectral clustering is not clear; even though the spectral clustering step can also have a critical impact on the overall performance. To resolve this issue, in this paper, we provide a nonparametric method to estimate the probabilistic cluster membership from these affinity matrices, which we can directly find the clusters from. The likelihood for an affinity matrix is defined nonparametrically based on histograms given the probabilistic membership, which is defined as a combination of probability simplices, and an additional prior probability is defined to regularize the membership as a Bernoulli distribution. Solving this maximum a posteriori problem replaces the spectral clustering procedure, and the final discrete cluster membership can be calculated by selecting the clusters with maximum probabilities. The proposed method provides state-of-the-art performance for well-known subspace clustering methods on popular benchmark databases.

Building a Compact Convolutional Neural Network for Embedded Intelligent Sensor Systems Using Group Sparsity and Knowledge Distillation.

As artificial intelligence (AI)- or deep-learning-based technologies become more popular,the main research interest in the field is not only on their accuracy, but also their efficiency, e.g., theability to give immediate results on the users' inputs. To achieve this, there have been many attemptsto embed deep learning technology on intelligent sensors. However, there are still many obstacles inembedding a deep network in sensors with limited resources. Most importantly, there is an apparenttrade-off between the complexity of a network and its processing time, and finding a structure witha better trade-off curve is vital for successful applications in intelligent sensors. In this paper, wepropose two strategies for designing a compact deep network that maintains the required level ofperformance even after minimizing the computations. The first strategy is to automatically determinethe number of parameters of a network by utilizing group sparsity and knowledge distillation (KD)in the training process. By doing so, KD can compensate for the possible losses in accuracy causedby enforcing sparsity. Nevertheless, a problem in applying the first strategy is the unclarity indetermining the balance between the accuracy improvement due to KD and the parameter reductionby sparse regularization. To handle this balancing problem, we propose a second strategy: a feedbackcontrol mechanism based on the proportional control theory. The feedback control logic determinesthe amount of emphasis to be put on network sparsity during training and is controlled based onthe comparative accuracy losses of the teacher and student models in the training. A surprising facthere is that this control scheme not only determines an appropriate trade-off point, but also improvesthe trade-off curve itself. The results of experiments on CIFAR-10, CIFAR-100, and ImageNet32 X 32datasets show that the proposed method is effective in building a compact network while preventingperformance degradation due to sparsity regularization much better than other baselines.

Variational Inference for 3-D Localization and Tracking of Multiple Targets Using Multiple Cameras.

This paper proposes a novel unified framework to solve the 3-D localization and tracking problem that occurs multiple camera settings with overlapping views. The main challenge is to overcome the uncertainty of the back projection arising from the challenges of ground point detection in an environment that includes severe occlusions and the unknown heights of people. To tackle this challenge, we establish a Bayesian learning framework that maximizes a posterior over the trajectory assignments and 3-D positions for given detections from multiple cameras. To solve the Bayesian learning problem in a tractable form, we develop an expectation-maximization scheme based on the variation inference approximation, where the probability distributions are designed to follow Boltzmann distributions of seven terms that are induced from multicamera tracking settings. The experimental results show that the proposed method outperforms the state-of-the-art methods on the challenging multicamera data sets.

Comparing phase- and amplitude-gated volumetric modulated arc therapy for stereotactic body radiation therapy using 3D printed lung phantom.

PURPOSE: To compare the dosimetric impact and treatment delivery efficacy of phase-gated volumetric modulated arc therapy (VMAT) vs amplitude-gated VMAT for stereotactic body radiation therapy (SBRT) for lung cancer by using realistic three-dimensional-printed phantoms. METHODS: Four patient-specific moving lung phantoms that closely simulate the heterogeneity of lung tissue and breathing patterns were fabricated with four planning computed tomography (CT) images for lung SBRT cases. The phantoms were designed to be bisected for the measurement of two-dimensional dose distributions by using EBT3 dosimetry film. The dosimetric accuracy of treatment under respiratory motion was analyzed with the gamma index (2%/1 mm) between the plan dose and film dose measured under phase- and amplitude-gated VMAT. For the validation of the direct usage of the real-time position management (RPM) data for respiratory motion, the relationship between the RPM signal and the diaphragm position was measured by four-dimensional CT. By using data recorded during the beam delivery of both phase- and amplitude-gated VMAT, the total time intervals were compared for each treatment mode. RESULTS: Film dosimetry showed a 5.2 ± 4.2% difference of gamma passing rate (2%/1 mm) on average between the phase- vs amplitude-gated VMAT [77.7% (72.7%-85.9%) for the phase mode and 82.9% (81.4%-86.2%) for the amplitude mode]. For delivery efficiency, frequent interruptions were observed during the phase-gated VMAT, which stopped the beam delivery and required a certain amount of time before resuming the beam. This abnormality in phase-gated VMAT caused a prolonged treatment delivery time of 366 s compared with 183 s for amplitude-gated VMAT. CONCLUSIONS: Considering the dosimetric accuracy and delivery efficacy between the gating methods, amplitude mode is superior to phase mode for gated VMAT treatment.

Face Recognition in SSPP Problem Using Face Relighting Based on Coupled Bilinear Model.

There have been decades of research on face recognition, and the performance of many state-of-the-art face recognition algorithms under well-conditioned environments has become saturated. Accordingly, recent research efforts have focused on difficult but practical challenges. One such issue is the single sample per person (SSPP) problem, i.e., the case where only one training image of each person. While this problem is challenging because it is difficult to establish the within-class variation, working toward its solution is very practical because often only a few images of a person are available. To address the SSPP problem, we propose an efficient coupled bilinear model that generates virtual images under various illuminations using a single input image. The proposed model is inspired by the knowledge that the illuminance of an image is not sensitive to the poor quality of a subspace-based model, and it has a strong correlation to the image itself. Accordingly, a coupled bilinear model was constructed that retrieves the illuminance information from an input image. This information is then combined with the input image to estimate the texture information, from which we can generate virtual illumination conditions. The proposed method can instantly generate numerous virtual images of good quality, and these images can then be utilized to train the feature space for resolving SSPP problems. Experimental results show that the proposed method outperforms the existing algorithms.

A single-optical kernel for a phosphor-screen-based geometric QA system (RavenQA™) as a tool for patient-specific IMRT/VMAT QA.

It has been proven that portal dosimetry can be derived from a mirror-based fluorescent EPID system by applying multiple kernels that are position dependent. The purpose of this study is to show that patient-specific IMRT/VMAT verification with a single kernel which is acquired from a series of output measurements of a few field sizes is feasible using a commercially available phosphor-screen-based geometric QA system. The optical scatter component in the RavenQA™ (LAP GmbH Laser Applications; Lüneberg, Germany) is corrected by deconvolution with a two-dimensional (2D) spatially invariant single optical scatter kernel (OSK). We assume that the OSK is a 2D isotropic point spread function that decreases as a function of distance from the scatter center. The OSK is determined by comparing output factors of various field sizes. We report on performance testing of the system using 12 intensity-modulated radiation therapy and three volumetric-modulated arc therapy cases. A single spatially invariant OSK can be employed, because the shapes of the OSK across the image plate are almost identical. The average 3%/3 mm gamma passing rate for 15 patients was 97.6% ± 1.1%. The passing rate was >95% for all patients. It is feasible to perform the patient-specific IMRT/VMAT verification with a single kernel using a commercially available phosphor-screen-based mechanical QA device in accordance with AAPM TG-142. It is also practical to implement since it only requires to measure the optical intensities of the field centers of several square fields, in order to obtain the OSK.

Loss of the E3 ubiquitin ligase MKRN1 represses diet-induced metabolic syndrome through AMPK activation.

AMP-activated protein kinase (AMPK) plays a key role in controlling energy metabolism in response to physiological and nutritional status. Although AMPK activation has been proposed as a promising molecular target for treating obesity and its related comorbidities, the use of pharmacological AMPK activators has been met with contradictory therapeutic challenges. Here we show a regulatory mechanism for AMPK through its ubiquitination and degradation by the E3 ubiquitin ligase makorin ring finger protein 1 (MKRN1). MKRN1 depletion promotes glucose consumption and suppresses lipid accumulation due to AMPK stabilisation and activation. Accordingly, MKRN1-null mice show chronic AMPK activation in both liver and adipose tissue, resulting in significant suppression of diet-induced metabolic syndrome. We demonstrate also its therapeutic effect by administering shRNA targeting MKRN1 into obese mice that reverses non-alcoholic fatty liver disease. We suggest that ubiquitin-dependent AMPK degradation represents a target therapeutic strategy for metabolic disorders.

Ablation of insulin receptor substrates 1 and 2 suppresses Kras-driven lung tumorigenesis.

Non-small-cell lung cancer (NSCLC) is a leading cause of cancer death worldwide, with 25% of cases harboring oncogenic Kirsten rat sarcoma (KRAS). Although KRAS direct binding to and activation of PI3K is required for KRAS-driven lung tumorigenesis, the contribution of insulin receptor (IR) and insulin-like growth factor 1 receptor (IGF1R) in the context of mutant KRAS remains controversial. Here, we provide genetic evidence that lung-specific dual ablation of insulin receptor substrates 1/2 (Irs1/Irs2), which mediate insulin and IGF1 signaling, strongly suppresses tumor initiation and dramatically extends the survival of a mouse model of lung cancer with Kras activation and p53 loss. Mice with Irs1/Irs2 loss eventually succumb to tumor burden, with tumor cells displaying suppressed Akt activation and strikingly diminished intracellular levels of essential amino acids. Acute loss of IRS1/IRS2 or inhibition of IR/IGF1R in KRAS-mutant human NSCLC cells decreases the uptake and lowers the intracellular levels of amino acids, while enhancing basal autophagy and sensitivity to autophagy and proteasome inhibitors. These findings demonstrate that insulin/IGF1 signaling is required for KRAS-mutant lung cancer initiation, and identify decreased amino acid levels as a metabolic vulnerability in tumor cells with IR/IGF1R inhibition. Consequently, combinatorial targeting of IR/IGF1R with autophagy or proteasome inhibitors may represent an effective therapeutic strategy in KRAS-mutant NSCLC.