Maximum A Posteriori Probability Research Articles

How do you know that you don’t know?

Whenever someone in a team tries to help others, it is crucial that they have some understanding of other team members’ goals. In modern teams, this applies equally to human and artificial (“bot”) assistants. Understanding when one does not know something is crucial for stopping the execution of inappropriate behavior and, ideally, attempting to learn more appropriate actions. From a statistical point of view, this can be translated to assessing whether none of the hypotheses in a considered set is correct. Here we investigate a novel approach for making this assessment based on monitoring the maximum a posteriori probability (MAP) of a set of candidate hypotheses as new observations arrive. Simulation studies suggest that this is a promising approach, however, we also caution that there may be cases where this is more challenging. The problem we study and the solution we propose are general, with applications well beyond human–bot teaming, including for example the scientific process of theory development.

Low-complexity and low-latency equalization technique - probabilistic noise cancellation.

Both inside data centers (DCs) and in short optical links between data centers (DC campuses), intensity-modulation and direct-detection (IMDD) systems using four-level pulse amplitude modulation (PAM4) will dominate this decade due to low transceiver price and power consumption. The next DC transceiver generation based on 100 Gbaud PAM4 will require advanced digital signal processing (DSP) algorithms and more powerful forward error correction (FEC) codes. Because of bandwidth limitations, the conventional DC DSP based on a few-tap linear feed-forward equalizer (FFE) is likely to be upgraded to more complex but still low-complexity Volterra equalizers followed by a noise whitening filter and either a maximum likelihood sequence estimation (MLSE) or a maximum a posteriori probability (MAP) algorithm. However, stringent power consumption and latency requirements may limit the use of complex algorithms such as decision feedback equalizer (DFE) or MLSE/MAP in DC networks (DCN). In this paper, we introduce a low-complexity, low-latency algorithm based on a feedforward structure, yielding a performance between DFE and MLSE. We call the novel equalization algorithm probabilistic noise cancellation (PNC), since it weights noise patterns based on their probabilities in the presence of bandwidth limitations. The probabilistic weighting is efficiently exploited in correcting correlated errors caused by noise coloring in the FFE.

A machine learning-based Bayesian model for predicting the duration of ship detention in PSC inspection

Port state control (PSC) inspections are deemed as an effective way to detect substandard vessels and ensure maritime safety around the world. Despite great efforts on PSC in recent years, one challenge that still exists in today’s PSC inspection practice is that there lack relevant scheme or academic research focusing on the duration of vessel detention, which is of great importance to the inspection system. To assist port authorities in estimating detention duration and minimizing the existence of substandard vessels, this paper aims to develop a novel data-driven machine learning based model based on the inspection records collected within the jurisdiction of Paris MoU from January 2015 to March 2022. The model is trained via the incorporation of an Improved Tree Augmented Naïve (ITAN) learning approach and a maximum a posteriori probability (MAP) of Expectation Maximization (EM) approach for the first time within the context of PSC research, which could be used as a prediction tool to determine rational durations for detained vessels. Thorough analysis of the proposed model enables the identification of risk variables and deficiency types having significant effects leading to long duration of detention. Further, the research findings could reveal managerial suggestions and insights for port authorities to reduce the occurrence of substandard vessels via the inspection system, i.e., identify specific risk level of vessels and ensure a more-efficient vessel selection process; design specific instructions and rules to regulate risk variables and deficiencies with huge effect on a long duration of detention. This research will provide insightful reference for effectively improving vessel quality, inspection efficiency, and maritime safety.

Application of General MAP Detection Theory for PAM Signals to the Analysis of TDECQ and SSPRQ

Transmitter and dispersion eye closure quaternary (TDECQ) has been selected by the IEEE Ethernet Standard as a vital test to enable the 4-level pulse amplitude modulation (PAM4) format for 400 Gb/s operation over fiber optic cables and short stress pattern random quaternary (SSPRQ) is exclusively elected as the testing signal for TDECQ and some other important tests. The authors find that the SSPRQ is an un-equiprobable sequence whereas the other PAM4 test patterns, such as PRBS15Q and PRBS31Q (or in general, PRBSnQ) are equiprobable. This motivates us to use a general maximum a posteriori probability (MAP) detection theory to handle the situation. We show the penalty using the SSPRQ signal, compared with the PRBSnQ, is unfortunately 0.135 dB in energy per bit to noise power density ratio (EbNo), coming from both the average energy decrease due to the un-equiprobability and the deviation from MAP detection at the given SER. The TDECQ tester with SSPRQ will predict a different BER compared with that using a BER tester.

Detection and ISI mitigation in mobile molecular communication system for targeted drug delivery

This paper proposes a targeted drug delivery system based on mobile molecular communication (MMC). The system consists of a mobile transmitter and a mobile reactive receiver. The transmitter can sense the required drug concentration and send commands to the receiver for drug delivery in the extracellular fluid (ECF). The commands are sent in the form of bits and the received signal is prone to noise and inter-symbol interference (ISI). Hence, at the receiver, two detection techniques, differential amplitude detector (DAD) and differential energy detector (DED) with ISI mitigation are proposed for MMC. Manchester-coded bits are transmitted using modified concentration shift keying (MCSK). In the proposed detection mechanism, an adaptive threshold technique is used for estimating the number of signaling molecules using the maximum a posteriori probability (MAP) rule. Further in each bit interval, dynamic distance estimation, signal reconstruction, and ISI mitigation are performed. Particle-based simulation for reactive receiver is also carried out to validate the results. A low bit error rate (BER) in the MMC system signifies the promising performance of the drug delivery system.

Codebook Mismatch can be Fully Compensated by Mismatched Decoding

We consider an ensemble of constant composition codes that are subsets of linear codes: while the encoder uses only the constant-composition subcode, the decoder operates as if the full linear code was used, with the motivation of simultaneously benefiting both from the probabilistic shaping of the channel input (to achieve higher rates) and from the linear structure of the code (to allow for lower complexity practical decoding). We prove that the codebook mismatch can be fully compensated by using a mismatched additive decoding metric that achieves the random coding error exponent of (non-linear) constant composition codes. As the coding rate tends to the mutual information, the optimal mismatched metric approaches the maximum a posteriori probability (MAP) metric, showing that codebook mismatch with MAP metric is capacity-achieving for the optimal input assignment.

Simultaneous Dual-Wavelength Source Raman Spectroscopy with a Handheld Confocal Probe for Analysis of the Chemical Composition of In Vivo Human Skin.

Confocal Raman spectroscopy (CRS) is a powerful tool that has been widely used for biological tissue analysis because of its noninvasive nature, high specificity, and rich biochemical information. However, current commercial CRS systems suffer from limited detection regions (450-1750 cm-1), bulky sizes, nonflexibilities, slow acquisitions by consecutive excitations, and high costs if using a Fourier transform (FT) Raman spectroscopy with an InGaAs detector, which impede their adoption in clinics. In this study, we developed a portable CRS system with a simultaneous dual-wavelength source and a miniaturized handheld probe (120 mm × 60 mm × 50 mm) that can acquire spectra in both fingerprint (FP, 450-1750 cm-1) and high wavenumber (HW, 2800-3800 cm-1) regions simultaneously. An innovative design combining 671 and 785 nm lasers for simultaneous excitation through a compact and high-efficiency (>90%) wavelength combiner was implemented. Moreover, to decouple the fused FP and HW spectra, a first-of-its-kind precise Raman spectra separation algorithm (PRSSA) was developed based on the maximum a posteriori probability (MAP) estimate. The accuracy of spectra separation was greater than 99%, demonstrated in both phantom experiments and in vivo human skin measurements. The total data acquisition time was reduced by greater than 50% compared to other CRS systems. The results proved our proposed CRS system and PRSSA's superior capability in fast and ultrawideband spectra acquisition will significantly improve the integration of CRS in the clinical workflow.

Development of Terahertz Imaging Markers for Pancreatic Ductal Adenocarcinoma using Maximum A Posteriori Probability (MAP) Estimation.

Pancreatic ductal adenocarcinoma (PDAC) is one of the significant reasons for cancer-related death in the United States due to a lack of timely prognosis and the poor efficacy of the standard treatment protocol. Immunotherapy-based neoadjuvant therapy, such as stereotactic body radiotherapy (SBRT), has shown promising results compared to conventional radiotherapy in strengthening the antitumor response in PDAC. To probe and quantify the antitumor response with SBRT, we propose to study the tumor microenvironment using terahertz time-domain spectroscopy (THz-TDS). Since the tumor's complex microenvironment plays a key role in disease progression and treatment supervision, THz-TDS can be a revolutionary tool to help in treatment planning by probing the changes in the tissue microenvironment. This paper presents THz-TDS of paraffin-embedded PDAC samples utilizing a clinically relevant genetically engineered mouse model. This Article aims to develop and validate a novel time-domain approximation method based on maximum a posteriori probability (MAP) estimation to extract terahertz parameters, namely, the refractive index and the absorption coefficient, from THz-TDS. Unlike the standard frequency-domain (FD) analysis, the parameters extracted from MAP construct better-conserved tissue parameters estimates, since the FD optimization often incorporates errors due to numerical instabilities during phase unwrapping, especially when propagating in lossy media. The THz-range index of refraction extracted from MAP and absorption coefficient parameters report a statistically significant distinction between PDAC tissue regions and their healthy equivalents. The coefficient of variation of the refractive index extracted by MAP is one order of magnitude lower compared to the one extracted from FD analysis. The index of refraction and absorption coefficient extracted from the MAP are suggested as the best imaging markers to reconstruct THz images of biological tissues to reflect their physical properties accurately and reproducibly. The obtained THz scans were validated using standard histopathology.

Probabilistic Structural Model Updating with Modal Flexibility Using a Modified Firefly Algorithm.

Structural model updating is one of the most important steps in structural health monitoring, which can achieve high-precision matching between finite element models and actual engineering structures. In this study, a Bayesian model updating method with modal flexibility was presented, where a modified heuristic optimization algorithm named modified Nelder-Mead firefly algorithm (m-NMFA) was proposed to find the most probable values (MPV) of model parameters for the maximum a posteriori probability (MAP) estimate. The proposed m-NMFA was compared to the original firefly algorithm (FA), the genetic algorithm (GA), and the particle swarm algorithm (PSO) through the numerical illustrative examples of 18 benchmark functions and a twelve-story shear frame model. Then, a six-story shear frame model test was performed to identify the inter-story stiffness of the structure in the original and the damage states, respectively. By comparing the two, the position and extent of damage were accurately found and quantified in a probabilistic manner. In terms of optimization, the proposed m-NMFA was powerful to find the MPVs much faster and more accurately. In the incomplete measurement case, only the m-NMFA achieved target damage identification results. The proposed Bayesian model updating method has the advantages of high precision, fast convergence, and strong robustness in MPV finding and the ability of parameter uncertainty quantification.

Dynamic dense CRF inference for video segmentation and semantic SLAM

The dense conditional random field (dense CRF) is an effective post-processing tool for image/video segmentation and semantic SLAM. In this paper, we extend the traditional dense CRF inference algorithm to incremental sensor data modelling. The algorithm efficiently infers the maximum a posteriori probability (MAP) solution for a dynamically changing dense CRF model that is applied to incremental multi-class video segmentation and semantic SLAM. The computational cost is roughly proportional to the total change in the Gaussian pairwise edges of the dense CRF. In our system, with an increase in the number of frames of the sensor data, MAP calculations take approximately the same time to compute the overall three-dimensional dense CRF modelled for the entire video. Compared with the traditional dense CRF for video segmentation, this method is more suitable for incremental (in-line) video segmentation and robot semantic SLAM. The results of experiments show that if part of a pairwise edge is altered, our dynamic algorithm is significantly faster than the widely known standard dense CRF algorithm. In addition, the accuracy of its inference does not change. Several multi-class video segmentation tests confirmed the efficiency of inference of the algorithm. In another application, we used the dynamic dense CRF to incrementally integrate robot SLAM and video segmentation. The results show that an accurate SLAM can improve the accuracy of video segmentation, and the computational cost of the dense CRF MAP can be constrained over a constant range. The application of our algorithm is not limited to video segmentation: It is generic, and can be used to yield similar improvements in many optimization solutions for MAP in dynamically changing models.

Variational inference for nonlinear inverse problems via neural net kernels: Comparison to Bayesian neural networks, application to topology optimization

Inverse problems and, in particular, inferring unknown or latent parameters from data are ubiquitous in engineering simulations. A predominant viewpoint in identifying unknown parameters is Bayesian inference where both prior information about the parameters and the information from the observations via likelihood evaluations are incorporated into the inference process. In this paper, we adopt a similar viewpoint with a slightly different numerical procedure from standard inference approaches to provide insight about the localized behavior of unknown underlying parameters. We present a variational inference approach which mainly incorporates the observation data in a point-wise manner, i.e. we invert a limited number of observation data leveraging the gradient information of the forward map with respect to parameters, and find true individual samples of the latent parameters when the forward map is noise-free and one-to-one. For statistical calculations (as the ultimate goal in simulations), a large number of samples are generated from a trained neural network which serves as a transport map from the prior to posterior latent parameters. Our neural network machinery, developed as part of the inference framework and referred to as Neural Net Kernels (NNK), is based on hierarchical (deep) kernels which provide greater flexibility for training compared to standard neural networks. We showcase the effectiveness of our inference procedure in identifying bimodal and irregular distributions compared to a number of approaches including a maximum a posteriori probability (MAP)-based approach, Markov Chain Monte Carlo sampling with both Metropolis–Hastings and Hamiltonian Monte Carlo algorithms, and a Bayesian neural network approach, namely the widely-known Bayes by Backprop algorithm via a pedagogical example. We further apply our inference procedure to two and three dimensional topology optimization problems where we identify the latent parameters in the random field elastic modulus, modeled as a Karhunen–Loéve expansion, considering the high dimensional design iterate-displacement pair as training data. As future research, we discuss the application of this inference approach in identifying constitutive models in nonlinear elasticity and development of fast linear algebraic solvers for large scale similar, i.e. finite element-based inverse problems calculations.

Bayesian statistics-guided label refurbishment mechanism: Mitigating label noise in medical image classification.

Deep neural networks (DNNs) have been widely applied in medical image classification, benefiting from its powerful mapping capability among medical images. However, these existing deep learning-based methods depend on an enormous amount of carefully labeled images. Meanwhile, noise is inevitably introduced in the labeling process, degrading the performance of models. Hence, it is significant to devise robust training strategies to mitigate label noise in the medical image classificationtasks. In this work, we propose a novel Bayesian statistics-guided label refurbishment mechanism (BLRM) for DNNs to prevent overfitting noisy images. BLRMutilizes maximum a posteriori probability in the Bayesian statistics and the exponentially time-weighted technique to selectively correct the labels of noisy images. The training images are purified gradually with the training epochs when BLRMis activated, further improving classificationperformance. Comprehensive experiments on both synthetic noisy images (public OCT & Messidor datasets) and real-world noisy images (ANIMAL-10N) demonstrate that BLRMrefurbishes the noisy labels selectively, curbing the adverse effects of noisy data. Also, the anti-noise BLRMsintegrated with DNNs are effective at different noise ratio and are independent of backbone DNN architectures. In addition, BLRMis superior to state-of-the-art comparative methods of anti-noise. These investigations indicate that the proposed BLRMis well capable of mitigating label noise in medical image classificationtasks.

Informing behaviour change intervention design using systematic review with Bayesian meta-analysis: physical activity in heart failure

ABSTRACT Embracing the Bayesian approach, we aimed to synthesise evidence regarding barriers and enablers to physical activity in adults with heart failure (HF) to inform behaviour change intervention. This approach helps estimate and quantify the uncertainty in the evidence and facilitates the synthesis of qualitative and quantitative studies. Qualitative evidence was annotated using the Theoretical Domains Framework and represented as a prior distribution using an expert elicitation task. The maximum a posteriori probability (MAP) for the probability distribution for the log OR was used to estimate the relationship between physical activity and each determinant according to qualitative, quantitative, and qualitative and quantitative evidence combined. The probability distribution dispersion (SD) was used to evaluate uncertainty in the evidence. Three qualitative and 16 quantitative studies were included (N = 2739). High pro-b-type natriuretic peptide (MAP = −1.16; 95%CrI: [−1.21; −1.11]) and self-reported symptoms (MAP = − 0.48; 95%CrI: [ −0.40; −0.55]) were suggested as barriers to physical activity with low uncertainty (SD = 0.18 and 0.19, respectively). Modifiable barriers were symptom distress (MAP = −0.46; 95%CrI: [−0.68; −0.24], SD = 0.36), and negative attitude (MAP = −0.40; 95%CrI: [−0.49; −0.31], SD = 0.26). Modifiable enablers were social support (MAP = 0.56; 95%CrI: [0.48; 0.63], SD = 0.26), self-efficacy (MAP = 0.43; 95%CrI: [0.32; 0.54], SD = 0.37), positive physical activity attitude (MAP = 0.92; 95%CrI: [0.77; 1.06], SD = 0.36).

Higher-Order Modulation vs Faster-Than-Nyquist PAM-4 for Datacenter IM-DD Optics: An AIR Comparison Under Practical Bandwidth Limits

The 4-ary pulse amplitude modulation (PAM-4) has been widely deployed to both electrical and optical interfaces in datacenter applications. Based on 100-GBd PAM-4, the 200G per lane optical interface is being actively developed in both industrial interoperability groups and standard bodies. Going beyond 200G, it is questionable if PAM-4 could support an even faster interface with higher baud-rate, considering the bandwidth growth for both electronics and optical components has started lagging the ever-increasing capacity demand. Under limited bandwidth, one option is to improve the spectral efficiency by higher order formats like PAM-6 and PAM-8, but this induces backward incompatibility as well as potential modulation mismatch between electrical and optical interfaces that requires additional energy for gearboxing. This revives the study of faster than Nyquist (FTN) signaling to pursue higher PAM-4 baud-rate breaking the bandwidth limit. In ECOC 2021, we demonstrated a reduced-complexity maximum a posteriori probability (MAP) decoder in intensity-modulation (IM) direct-detection (DD) systems with superior FTN performance. In this paper, we extend our work by adding more algorithm details and experimental results. Moreover, we perform the achievable information rate (AIR) comparison between the FTN PAM-4 and the higher-order modulation within the Nyquist limit in different channel conditions. Remarkably, in a practical IM-DD channel with gradually decayed frequency response, FTN shows AIR gain over the Nyquist approach even at a low FTN decoding complexity. The finding suggests FTN PAM could be a promising candidate at post 200G era to offer both AIR and implementation advantages.

GMM based low-complexity adaptive machine-learning equalizers for optical fiber communication

The demand for high-speed data transmission has increased rapidly over the past few years, leading to advanced optical communication techniques. However, most of machine learning based equalizers which are trained with offline dataset mainly focus on achieving low BER, neglecting the generalization ability when the properties of optical link change. In this paper, we propose Gaussian Mixture model (GMM) based low-complexity adaptive machine-learning equalizers. The proposed online training strategy can fine-tune the parameters in GMM with a small amount of training sequence by introducing Maximum a posteriori probability (MAP) algorithm and sliding window. The size of online training sequence can be reduced by utilizing the parameters trained from offline data as the priori information to reduce the size of online training sequence. The proposed adaptive GMM based equalizers can not only show an excellent capability of mitigating nonlinear distortions but also show the ability to update parameters automatically according to the channel conditions. In addition, experimental results show that by introducing MAP algorithm and sliding window, the convergence speed of the proposed equalizers have been accelerated by 2 times, compared with the case that the traditional Expectation Maximization (EM) algorithm is used for online parameter update. On the other hand, there is no significant increase in computational complexity during the online stage.

Joint ML/MAP Estimation of the Frequency and Phase of a Single Sinusoid With Wiener Carrier Phase Noise

We address here the issue of jointly estimating the angle parameters of a single sinusoid with Wiener carrier phase noise and observed in additive, white, Gaussian noise (AWGN). We develop the theoretical foundation for time-domain, phase-based, joint maximum likelihood (ML) estimation of the unknown carrier frequency and the initial carrier phase, with simultaneous maximum a posteriori probability (MAP) estimation of the time-varying carrier phase noise. The derivation is based on the amplitude and phase-form of the noisy received signal model together with the use of the best, linearized, additive observation phase noise model due to AWGN. Our newly derived estimators are closed-form expressions, consisting of both the phase and the magnitude of all the received signal samples. More importantly, they all have a low-complexity, sample-by-sample iterative processing structure, which can be implemented iteratively in real-time. As a basis for comparison, the Cramer-Rao lower bound (CRLB) for the ML estimators and the Bayesian CRLB (BCRLB) for the MAP estimator are derived in the presence of carrier phase noise, and the results simply depend on the signal-to-noise ratio (SNR), the observation length and the phase noise variance. It is theoretically shown that the estimates obtained are unbiased, and the mean-square error (MSE) of the estimators attain the CRLB/BCRLB at high SNR. The MSE performance as a function of the SNR, the observation length and the phase noise variance is verified using Monte Carlo simulation, which shows a remarkable improvement in estimation accuracy in large phase noise.

Signal Segmentation Using Maximum a Posteriori Probability Estimator with Application in Artifact ”corrected” EEG Data

The ”corrected” EEG data, after artifact removing, may be the subject of further investigations, for example segmentation, result- ing new information to be used for feature ex- traction, of great help for medical diagnosis. The paper has as object a generally approach for seg- mentation, making use of Maximum A posteriori Probability (MAP) estimator. The proposed pro- cedure has been used in the analysis of a sample lowpass EEG signals recorded with 13 scalp and 1 EOG electrodes, event-related potential (ERP) data.

An Efficient Turbo Decoding and Frequency Domain Turbo Equalization for LTE Based Narrowband Internet of Things (NB-IoT) Systems.

This paper addresses the main crucial aspects of physical (PHY) layer channel coding in uplink NB-IoT systems. In uplink NB-IoT systems, various channel coding algorithms are deployed due to the nature of the adopted Long-Term Evolution (LTE) channel coding which presents a great challenge at the expense of high decoding complexity, power consumption, error floor phenomena, while experiencing performance degradation for short block lengths. For this reason, such a design considerably increases the overall system complexity, which is difficult to implement. Therefore, the existing LTE turbo codes are not recommended in NB-IoT systems and, hence, new channel coding algorithms need to be employed for LPWA specifications. First, LTE-based turbo decoding and frequency-domain turbo equalization algorithms are proposed, modifying the simplified maximum a posteriori probability (MAP) decoder and minimum mean square error (MMSE) Turbo equalization algorithms were appended to different Narrowband Physical Uplink Shared Channel (NPUSCH) subcarriers for interference cancellation. These proposed methods aim to minimize the complexity of realizing the traditional MAP turbo decoder and MMSE estimators in the newly NB-IoT PHY layer features. We compare the system performance in terms of block error rate (BLER) and computational complexity.

Multi-fidelity Bayesian neural networks: Algorithms and applications

We propose a new class of Bayesian neural networks (BNNs) that can be trained using noisy data of variable fidelity, and we apply them to learn function approximations as well as to solve inverse problems based on partial differential equations (PDEs). These multi-fidelity BNNs consist of three neural networks: The first is a fully connected neural network, which is trained following the maximum a posteriori probability (MAP) method to fit the low-fidelity data; the second is a Bayesian neural network employed to capture the cross-correlation with uncertainty quantification between the low- and high-fidelity data; and the last one is the physics-informed neural network, which encodes the physical laws described by PDEs. For the training of the last two neural networks, we first employ the mean-field variational inference (VI) to maximize the evidence lower bound (ELBO) to obtain informative prior distributions for the hyperparameters in the BNNs, and subsequently we use the Hamiltonian Monte Carlo (HMC) method to estimate accurately the posterior distributions for the corresponding hyperparameters. We demonstrate the accuracy of the present method using synthetic data as well as real measurements. Specifically, we first approximate a one- and four-dimensional function, and then infer the reaction rates in one- and two-dimensional diffusion-reaction systems. Moreover, we infer the sea surface temperature (SST) in the Massachusetts and Cape Cod Bays using satellite images and in-situ measurements. Taken together, our results demonstrate that the present method can capture both linear and nonlinear correlation between the low- and high-fidelity data adaptively, identify unknown parameters in PDEs, and quantify uncertainties in predictions, given a few scattered noisy high-fidelity data. Finally, we demonstrate that we can effectively and efficiently reduce the uncertainties and hence enhance the prediction accuracy with an active learning approach, using as examples a specific one-dimensional function approximation and an inverse PDE problem.

Detecting the Early Infarct Core on Non-Contrast CT Images with a Deep Learning Residual Network

PurposeTo explore a new approach mainly based on deep learning residual network (ResNet) to detect infarct cores on non-contrast CT images and improve the accuracy of acute ischemic stroke diagnosis. MethodsWe continuously enrolled magnetic resonance diffusion weighted image (MR-DWI) confirmed first-episode ischemic stroke patients (onset time: less than 9 h) as well as some normal individuals in this study. They all underwent CT plain scan and MR-DWI scan with same scanning range, layer thickness (4 mm) and interlayer spacing (4 mm) (The time interval between two examinations: less than 4 h). Setting MR-DWI as gold standard of infarct core and using deep learning ResNet combined with a maximum a posteriori probability (MAP) model and a post-processing method to detect the infarct core on non-contrast CT images. After that, we use decision curve analysis (DCA) establishing models to analyze the value of this new method in clinical practice. Results116 ischemic stroke patients and 26 normal people were enrolled. 58 patients were allocated into training dataset and 58 were divided into testing dataset along with 26 normal samples. The identification accuracy of our ResNet based approach in detecting the infarct core on non-contrast CT is 75.9%. The DCA shows that this deep learning method is capable of improving the net benefit of ischemic stroke patients. ConclusionsOur deep learning residual network assisted with optimization methods is able to detect early infarct core on non-contrast CT images and has the potential to help physicians improve diagnostic accuracy in acute ischemic stroke patients.