The Usefulness of Serum Interleukin-6 as a Predictor of Response to Atezolizumab plus Bevacizumab Combination Treatment in Hepatocellular Carcinoma.

INTRODUCTION: In atezolizumab plus bevacizumab (Atezo/Bev) combination treatment, both drugs act on the immune system. Previously, we reported that immunological changes after Atezo/Bev administration for unresectable hepatocellular carcinoma (uHCC) revealed significant alterations in interleukin (IL)-6, soluble IL-2 receptor, tumor necrosis factor-alpha, and programmed cell death-1 levels. Among these variable factors, serum levels of IL-6 can be easily measured on a commercial basis. Therefore, this study aimed to investigate the utility of serum IL-6 as a predictor of tumor response to Atezo/Bev treatment for uHCC. METHOD: The study included 44 patients with HCC treated with Atezo/Bev. Blood samples were collected before and 3 weeks after treatment, and tumor response was assessed using contrast-enhanced computed tomography 6 weeks after treatment. RESULTS: Significant changes in serum IL-6 levels were observed in patients treated with Atezo/Bev as first-line therapy but not in those treated with it as second-line or later-line therapy. In patients treated with Atezo/Bev as first-line therapy, serum IL-6 levels increased significantly after treatment in patients with a complete or partial response but not in patients with stable or progressive disease. Furthermore, compared to other tumor markers such as alpha-fetoprotein, Lens culinaris agglutinin-reactive fraction of alpha-fetoprotein, and des-gamma-carboxyprothrombin, serum IL-6 levels exhibited the highest sensitivity in predicting tumor response during the treatment period. CONCLUSION: In patients with uHCC treated with Atezo/Bev, serum IL-6 levels could serve as a potential predictor of tumor response. Elevated levels after treatment may indicate a favorable tumor response and prognosis.

Variceal bleeding following treatment with atezolizumab plus bevacizumab in two patients with unresectable hepatocellular carcinoma.

Bleeding-related adverse events may occur due to anti-vascular endothelial growth factors. Here, we report two cases of variceal rupture during atezolizumab plus bevacizumab (ATZ/BV) treatment for unresectable hepatocellular carcinoma (u-HCC).Case 1 involved a man in his 60 s with alcoholic liver cirrhosis (LC) and u-HCC. Seventy-four days after ATZ/BV administration, the patient was admitted for hematemesis. Upper esophagogastroduodenoscopy (EGD) revealed worsening of the esophageal varices (EVs) to F2 grade with active bleeding. Endoscopic variceal ligation successfully achieved hemostasis.Case 2 involved a man in his 70 s with alcoholic LC and u-HCC. The patient was admitted with hematemesis 114 days after ATZ/BV administration. During EGD, the EVs deteriorated to F3 grade, although hemostasis had already been achieved. The evaluation was discontinued during the observation stage because of the worsening hepatic reserve.Neither patient had EVs warranting prophylactic treatment before ATZ/BV administration, showed a partial tumor response, or had portal vein tumor thrombus. Both patients demonstrated increased total diameters of the collateral veins and splenic volume compared to those before treatment. These findings suggest that ATZ/BV treatment may increase portal pressure. In conclusion, the administration of ATZ/BV to patients with LC and u-HCC necessitates careful management of EVs aggravation and rupture.

Healthy dietary choices involve prefrontal mechanisms associated with long-term reward maximization but not working memory.

Taste and health are critical factors to be considered when choosing foods. Prioritizing healthiness over tastiness requires self-control. It has also been suggested that self-control is guided by cognitive control. We then hypothesized that neural mechanisms underlying healthy food choice are associated with both self-control and cognitive control. Human participants performed a food choice task and a working memory task during functional MRI scanning. Their degree of self-control was assessed behaviorally by the value discount of delayed monetary rewards in intertemporal choice. Prioritizing healthiness in food choice was associated with greater activity in the superior, dorsolateral, and medial prefrontal cortices. Importantly, the prefrontal activity was greater in individuals with smaller delay discounting (i.e. high self-control) who preferred a delayed larger reward to an immediate smaller reward in intertemporal choice. On the other hand, working memory activity did not show a correlation with delay discounting or food choice activity, which was further supported by supplementary results that analyzed data from the Human Connectome Project. Our results suggest that the prefrontal cortex plays a critical role in healthy food choice, which requires self-control, but not working memory, for maximization of reward attainments in a remote future.

Location analysis of presynaptically active and silent synapses in single-cultured hippocampal neurons.

A morphologically present but non-functioning synapse is termed a silent synapse. Silent synapses are categorized into "postsynaptically silent synapses," where AMPA receptors are either absent or non-functional, and "presynaptically silent synapses," where neurotransmitters cannot be released from nerve terminals. The presence of presynaptically silent synapses remains enigmatic, and their physiological significance is highly intriguing. In this study, we examined the distribution and developmental changes of presynaptically active and silent synapses in individual neurons. Our findings show a gradual increase in the number of excitatory synapses, along with a corresponding decrease in the percentage of presynaptically silent synapses during neuronal development. To pinpoint the distribution of presynaptically active and silent synapses, i.e., their positional information, we employed Sholl analysis. Our results indicate that the distribution of presynaptically silent synapses within a single neuron does not exhibit a distinct pattern during synapse development in different distance from the cell body. However, irrespective of neuronal development, the proportion of presynaptically silent synapses tends to rise as the projection site moves farther from the cell body, suggesting that synapses near the cell body may exhibit higher synaptic transmission efficiency. This study represents the first observation of changes in the distribution of presynaptically active and silent synapses within a single neuron.

Remaining Issues Related to Serum Cytokines in Patients with Unresectable Hepatocellular Carcinoma Treated by Atezolizumab plus Bevacizumab Combination Treatment.

INTRODUCTION: Atezolizumab plus bevacizumab (AteBev) combination treatment is widely used as first-line systemic therapy for unresectable hepatocellular carcinoma (uHCC). We aimed to clarify therapeutic issues regarding serum cytokines and the immune reaction in patients with uHCC treated with AteBev. METHODS: We analyzed preserved serum from a previous prospective study on adult Japanese patients with chronic liver disease and uHCC who received AteBev treatment at our hospital. In that study, AteBev was administered intravenously every 3 weeks, and blood samples were collected before and after 3 weeks' treatment. Dynamic computed tomography was performed after 6 weeks of treatment to assess response. RESULTS: In the prospective study, 21 of the 59 patients showed partial response (PR) and 19 patients showed stable disease, but 19 patients showed progressive disease (PD). We found that serum levels of tumor necrosis factor-alpha, interleukin (IL)-6, and soluble IL-2 receptor (IL-2R) increased significantly in the PR group, but only soluble IL-2R increased significantly in the PD group. Regulatory T cells decreased significantly in the PD group, but there was no significant change in Th1 or Th2 cells from before to after treatment in any group. As regards soluble MHC-class I, pre-treatment levels were significantly lower in the PD group than in the PR group, and serum levels increased significantly with treatment in the PD group. CONCLUSION: These findings reveal a need to further improve T-cell priming and to further make T cells recognize tumor antigens in uHCC.

Evaluation of the Hierarchical Correspondence between the Human Brain and Artificial Neural Networks: A Review.

Artificial neural networks (ANNs) that are heavily inspired by the human brain now achieve human-level performance across multiple task domains. ANNs have thus drawn attention in neuroscience, raising the possibility of providing a framework for understanding the information encoded in the human brain. However, the correspondence between ANNs and the brain cannot be measured directly. They differ in outputs and substrates, neurons vastly outnumber their ANN analogs (i.e., nodes), and the key algorithm responsible for most of modern ANN training (i.e., backpropagation) is likely absent from the brain. Neuroscientists have thus taken a variety of approaches to examine the similarity between the brain and ANNs at multiple levels of their information hierarchy. This review provides an overview of the currently available approaches and their limitations for evaluating brain-ANN correspondence.

Importance of Atezolizumab Plus Bevacizumab Combination Treatment as First-line Therapy for Immunological Changes in Patients With Unresectable Hepatocellular Carcinoma.

BACKGROUND/AIM: The atezolizumab plus bevacizumab (AteBev) therapy is recommended as first-line treatment for unresectable hepatocellular carcinoma (uHCC). However, there remains a need to examine its efficacy with and without previous chemotherapy. Therefore, in patients with uHCC who underwent AteBev therapy, we aimed to clarify the effects of previous chemotherapy by examining serum immunological changes. PATIENTS AND METHODS: We retrospectively analyzed data of 29 patients with uHCC treated by AteBev therapy as part of a prospective study and divided participants into two groups depending on whether they had received prior chemotherapy. Dynamic computed tomography was performed after 6 weeks of treatment. Blood samples were collected at baseline and after 3 weeks of treatment. RESULTS: The group with prior treatment included 15 patients and the group without prior treatment included 14 patients. Objective response rates after six weeks of treatment were 13.3% and 28.6% in the groups with and without prior treatment, respectively. Serum levels of interleukin (IL)-6 and tumor necrosis factor-alpha showed no significant change in the group with prior treatment but increased significantly in the group without prior treatment. The percentage of regulatory T cells decreased significantly after treatment only in the group without prior treatment. CONCLUSION: In patients with uHCC, AteBev therapy can be expected to elicit an effective immune response in patients without prior treatment, but it may not do so in patients with prior treatment. Thus, AteBev appears to be more effective when used as first-line chemotherapy.

Ketamine-induced 1-Hz oscillation of spontaneous neural activity is not directly visible in the hemodynamics.

The extent to which resting-state hemodynamics reflects the underlying neural activity is still under debate. Especially in the delta frequency band (0.5-4 Hz), it is unclear whether the hemodynamics can directly track the dynamics of underlying neural activity. Based on a recent report showing that ketamine administration induced a 1-Hz neural activity oscillation in the retrosplenial cortex, we conducted simultaneous recordings of the calcium signal and hemodynamics in mice and examined whether the hemodynamics tracked the oscillatory neural activity. Although we observed that the oscillation induced by ketamine appeared in the calcium signal, no sign of oscillation was detected in the simultaneously recorded hemodynamics. Consistently, there was a notable decrease in the correlation between simultaneously recorded calcium signal and hemodynamics. However, on a much longer time scale (10-60 min), we unexpectedly observed an ultraslow increase of hemodynamic signals specifically in the same cortical region exhibiting the neural activity oscillation. These results indicated that hemodynamics cannot track the 1-Hz oscillation in neural activity, although the presence of neural activity oscillation was detectable on a longer timescale. Such ultraslow hemodynamics may be useful for detecting abnormal neural activity induced by psychotic drugs or mental disorders.

Effects of Lenvatinib on Skeletal Muscle Volume and Cardiac Function in Patients with Hepatocellular Carcinoma.

INTRODUCTION: Previously, we reported that the tyrosine kinase inhibitor (TKI) sorafenib decreases serum levels of carnitine and reduces skeletal muscle volume. Moreover, others reported that TKIs might lead to cardiomyopathy or heart failure. Therefore, this study aimed to evaluate the effects of lenvatinib (LEN) on skeletal muscle volume and cardiac function in patients with hepatocellular carcinoma (HCC). METHODS: This retrospective study included 58 adult Japanese patients with chronic liver diseases and HCC treated with LEN. Blood samples were collected before and after 4 weeks of treatment, and serum carnitine fraction and myostatin levels were measured. Before and after 4-6 weeks of treatment, the skeletal muscle index (SMI) was evaluated from computed tomography images and cardiac function was assessed by ultrasound cardiography. RESULTS: After treatment, SMI, serum levels of total carnitine, and global longitudinal strain were significantly lower, but serum levels of myostatin were significantly higher. Left ventricular ejection fraction showed no significant change. CONCLUSION: In patients with HCC, LEN decreases serum levels of carnitine, skeletal muscle volume, and worsens cardiac function.

Static and Dynamic Functional Connectivity Alterations in Alzheimer's Disease and Neuropsychiatric Diseases.

Background: To date, numerous studies have documented various alterations in resting brain activity in Alzheimer's disease (AD) and other neuropsychiatric diseases. In particular, disease-related alterations of functional connectivity (FC) in the resting state networks (RSN) have been documented. Altered FC in RSN is useful not only for interpreting the phenotype of diseases but also for diagnosing the diseases. More recently, several studies proposed the dynamics of resting-brain activity as a useful marker for detecting altered RSNs related to AD and other diseases. Objectives: In this article, we review recent studies exploring alterations of static and dynamic functional connectivity in AD and other neuropsychiatric diseases. We then discuss how to utilize and interpret dynamics of FC for studying resting brain activity in diseases. Results: In contrast to previous studies, which focused on FC calculated using an entire fMRI scan (static FC), newer studies focused on the temporal dynamics of FC within the scan (dynamic FC) to provide more sensitive measures to characterize RSNs. However, despite the increasing popularity of dynamic FC, several statistical investigations of dynamic FC cautioned that the results obtained in commonly used analyses for dynamic FC require careful interpretation. Conclusions: Although static and dynamic FC are likely to be a useful tool to detect altered RSN in patients affected by AD and other neuropsychiatric disorders, interpretation of altered dynamic FC in patients require special care. Impact statement We review recent studies of static and dynamic functional connectivity (dFC) in Alzheimer's disease and discuss interpretation of dFC.

Lenvatinib Might Induce Activation of Host Immunity in Patients with Hepatocellular Carcinoma.

INTRODUCTION: Atezolizumab, an immune checkpoint inhibitor, plus bevacizumab, a monoclonal antibody that binds to vascular endothelial growth factor (VEGF), is an approved first-line systemic treatment for unresectable hepatocellular carcinoma (HCC). Immune checkpoint inhibitors are more effective in patients with HCC when administered with anti-VEGF drugs; however, these drugs affect host immunity. Lenvatinib is an anti-VEGF agent used to treat HCC; therefore, this study evaluated the effect of treatment of HCC with lenvatinib on host immunity in patients with chronic liver disease (CLD). METHODS: We studied adult Japanese patients with CLD and unresectable HCC treated with lenvatinib at our hospital. Lenvatinib was administered for 4 weeks (8 mg/day for bodyweight <60 kg; 12 mg/day for bodyweight >60 kg). Blood samples were collected at baseline and at 4 weeks of treatment and examined for immune-related changes. RESULTS: Forty-three patients were enrolled in this study. We found a significant increase in T helper (Th) 1 cells following 4 weeks of lenvatinib treatment, although there was no significant difference in Th2 cells and regulatory T cells. We also found a significant increase in serum levels of TNF-alpha, soluble TNF-alpha receptor I, and endothelial growth factor following 4 weeks of lenvatinib treatment. Furthermore, an increase in Th1 cells and serum levels of TNF-alpha was found in patients with partial response. CONCLUSION: Lenvatinib might induce Th1-dominant host immunity in patients with CLD and unresectable HCC treatment in patients who showed a partial response. These changes in host immunity may be a biomarker in HCC patients treated with lenvatinib.

Measurement of skeletal muscle volume is useful for predicting prognosis in patients with liver cirrhosis.

BACKGROUND: Hepatic venous pressure gradient (HVPG) and the model for end-stage liver disease (MELD) score were previously reported as prognostic factors for outcome in patients with liver cirrhosis (LC), and recently, the presence of sarcopenia was reported to be an indicator of worse prognosis in these patients. AIM: This retrospective study aimed to clarify the importance of sarcopenia as a prognostic factor in patients with LC. MATERIAL AND METHODS: The MELD-Na score, HVPG, and skeletal muscle index (SMI) were measured in 202 patients between January 2013 and August 2020. We performed linear regression analysis between HVPG and SMI and calculated suitable cutoff values of HVPG for predicting presarcopenia and of HVPG, ΔSMI (i.e. the decrease in SMI per year, for predicting survival). Overall survival rates with the HVPG and ΔSMI cutoff values were compared by Kaplan-Meier estimates and log-rank tests. Prognostic factors for survival were analyzed by Cox regression univariate and multivariate analyses. RESULTS: In total, 71% (143/202) of patients presented with presarcopenia. Linear regression showed a significantly negative correlation between HVPG and SMI. Survival was significantly worse in the group with presarcopenia than in the group without. Survival was worse also in the group with an HVPG value ≥ 15 and ΔSMI ≥ -2.4. Cox regression multivariate analyses showed that MELD-Na score, HVPG, HVPG ≥ 15, ΔSMI, and ΔSMI ≥ -2.4 were independent prognostic factors. CONCLUSION: Skeletal muscle volume, especially ΔSMI, has a prognostic value equivalent to that of the MELD-Na score and HVPG.

Modular strategy for development of the hierarchical visual network in mice.

Hierarchical and parallel networks are fundamental structures of the mammalian brain1-8. During development, lower- and higher-order thalamic nuclei and many cortical areas in the visual system form interareal connections and build hierarchical dorsal and ventral streams9-13. One hypothesis for the development of visual network wiring involves a sequential strategy wherein neural connections are sequentially formed alongside hierarchical structures from lower to higher areas14-17. However, this sequential strategy would be inefficient for building the entire visual network comprising numerous interareal connections. We show that neural pathways from the mouse retina to primary visual cortex (V1) or dorsal/ventral higher visual areas (HVAs) through lower- or higher-order thalamic nuclei form as parallel modules before corticocortical connections. Subsequently, corticocortical connections among V1 and HVAs emerge to combine these modules. Retina-derived activity propagating the initial parallel modules is necessary to establish retinotopic inter-module connections. Thus, the visual network develops in a modular manner involving initial establishment of parallel modules and their subsequent concatenation. Findings in this study raise the possibility that parallel modules from higher-order thalamic nuclei to HVAs act as templates for cortical ventral and dorsal streams and suggest that the brain has an efficient strategy for the development of a hierarchical network comprising numerous areas.

Usefulness of partial splenic embolization for left-sided portal hypertension in a patient with a pancreatic neuroendocrine neoplasm: a case report and review of the literature.

Left-side portal hypertension (LSPH) is caused by isolated obstruction of the splenic vein and is associated with esophagogastric varices that extend from the lower esophagus to the greater curvature of the gastric body. Here, we report on a 74-year-old man with a pancreatic neuroendocrine neoplasm (NEN) in the pancreatic tail with multiple liver metastases. We decided that partial splenic embolization (PSE) was the best course of treatment to prevent rupture of the gastric varices, which were classified as markedly enlarged, nodular, or tumor-shaped and showed erosion of the mucosa. After PSE, the patient had no major complications and was discharged. At 3 and 6 months after the procedure, esophagogastroduodenoscopy and enhanced computerized tomography showed that the gastric varices had improved. This case demonstrates the usefulness of PSE for LSPH in patients with unresected pancreatic NEN.

On co-activation pattern analysis and non-stationarity of resting brain activity.

The non-stationarity of resting-state brain activity has received increasing attention in recent years. Functional connectivity (FC) analysis with short sliding windows and coactivation pattern (CAP) analysis are two widely used methods for assessing the dynamic characteristics of brain activity observed with functional magnetic resonance imaging (fMRI). However, the statistical nature of the dynamics captured by these techniques needs to be verified. In this study, we found that the results of CAP analysis were similar for real fMRI data and simulated stationary data with matching covariance structures and spectral contents. We also found that, for both the real and simulated data, CAPs were clustered into spatially heterogeneous modules. Moreover, for each of the modules in the real data, a spatially similar module was found in the simulated data. The present results suggest that care needs to be taken when interpreting observations drawn from CAP analysis as it does not necessarily reflect non-stationarity or a mixture of states in resting brain activity.

Executive control by fronto-parietal activity explains counterintuitive decision behavior in complex value-based decision-making.

In real life, humans make decisions by taking into account multiple independent factors, such as delay and probability. Cognitive psychology suggests that cognitive control mechanisms play a key role when facing such complex task conditions. However, in value-based decision-making, it still remains unclear to what extent cognitive control mechanisms become essential when the task condition is complex. In this study, we investigated decision-making behaviors and underlying neural mechanisms using a multifactor gambling task where participants simultaneously considered probability and delay. Decision-making behavior in the multifactor task was modulated by both probability and delay. The behavioral effect of probability was stronger than delay, consistent with previous studies. Furthermore, in a subset of conditions that recruited fronto-parietal activations, reaction times were paradoxically elongated despite lower probabilistic uncertainty. Notably, such a reaction time elongation did not occur in control tasks involving single factors. Meta-analysis of brain activations suggested an interpretation that the paradoxical increase of reaction time may be associated with strategy switching. Consistent with this interpretation, logistic regression analysis of the behavioral data suggested a presence of multiple decision strategies. Taken together, we found that a novel complex value-based decision-making task cause prominent activations in fronto-parietal cortex. Furthermore, we propose that these activations can be interpreted as recruitment of cognitive control system in complex situations.

A method for reconstruction of interpretable brain networks from transient synchronization in resting-state BOLD fluctuations.

Resting-state (rs) fMRI has been widely used to examine brain-wide large-scale spatiotemporal architectures, known as resting-state networks (RSNs). Recent studies have focused on the temporally evolving characteristics of RSNs, but it is unclear what temporal characteristics are reflected in the networks. To address this issue, we devised a novel method for voxel-based visualization of spatiotemporal characteristics of rs-fMRI with a time scale of tens of seconds. We first extracted clusters of dominant activity-patterns using a region-of-interest approach and then used these temporal patterns of the clusters to obtain voxel-based activation patterns related to the clusters. We found that activation patterns related to the clusters temporally evolved with a characteristic temporal structure and showed mutual temporal alternations over minutes. The voxel-based representation allowed the decoding of activation patterns of the clusters in rs-fMRI using a meta-analysis of functional activations. The activation patterns of the clusters were correlated with behavioral measures. Taken together, our analysis highlights a novel approach to examine brain activity dynamics during rest.

Adaptation of lenvatinib treatment in patients with hepatocellular carcinoma and portal vein tumor thrombosis.

PURPOSE: The aim of this study was to clarify the adaptation of lenvatinib treatment in patients with hepatocellular carcinoma (HCC) and portal vein tumor thrombosis (PVTT). METHOD: Fifty-three patients with HCC were treated with lenvatinib. Before and after treatment blood sampling, patients were examined by computed tomography and ultrasonography. In patients with portal trunk invasion (Vp4), the analysis focused on the degree of occlusion due to the tumor in the portal trunk. In patients without major PVTT {ie, invasion of the primary branch of the portal vein [Vp3] or Vp4}, portal blood flow volume was measured by Doppler analysis; however, Doppler analysis is difficult to perform in patients with major PVTT, so the time from administration of the contrast agent to when it reached the primary branch of the portal vein (portal vein arrival time) was evaluated with the contrast agent Sonazoid. RESULTS: Patients with Vp4 had a significantly worse prognosis than patients with Vp3 and a significant increase in Child-Pugh score at 2 months. Patients with major PVTT had a poor prognosis if the degree of occlusion of the portal trunk was 70% or more. In patients without major PVTT, portal blood flow was significantly decreased after administration of lenvatinib; and in patients with major PVTT, the hepatic artery and portal vein arrival times were significantly increased. CONCLUSION: Lenvatinib treatment should be avoided in patients with Vp4 with a high degree of portal trunk occlusion because of concerns about decreased portal blood flow.

Usefulness of virtual touch tissue quantification for predicting the presence of esophageal varices in patients with liver cirrhosis.

BACKGROUND AND AIM: Measuring the hepatic venous pressure gradient (HVPG) is an established technique to detect increased portal pressure and predict the presence of esophageal varices (EVs); however, the risk of the test is greater than the information it provides. This study aimed to clarify the usefulness of virtual touch tissue quantification (VTQ), which assesses liver stiffness, in predicting the presence of EVs in patients with liver cirrhosis by comparing it with HVPG. METHODS: Two hundred seventeen patients with liver cirrhosis underwent VTQ, HVPG measurement, and upper endoscopy. Patients were divided into three groups: group V, hepatitis C virus liver cirrhosis (n = 40); group A, alcoholic liver cirrhosis (n = 116); and group N, other liver cirrhosis (n = 61). In each group, we performed linear regression analysis of VTQ and HVPG data. The accuracy of VTQ and HVPG measurement in predicting the presence of EVs and high-risk EVs (EV category F2 and F3) was assessed by area under the receiver operating characteristic curve (AUROC). RESULTS: VTQ was significantly correlated with the HVPG in the whole patients and in each group, and both VTQ and HVPG values were significantly higher in patients with EVs and high-risk EVs than in those without. The AUROC for the presence of EVs for VTQ was 0.76 in the whole sample, 0.76 in group V, 0.79 in group A, and 0.67 in group N; and for HVPG, 0.92, 0.94, 0.93, and 0.88, respectively. For VTQ, the AUROC for the presence of high-risk EVs was 0.78 in the whole sample, 0.78 in group V, 0.73 in group A, and 0.73 in group N; and for HVPG, it was 0.85, 0.82, 0.85, and 0.82, respectively. CONCLUSION: VTQ was reliable at predicting the presence of EVs and high-risk EVs. Therefore, we propose that VTQ is a useful, noninvasive tool for predicting the presence of EVs in daily medical care.

Counterfactual Explanation of Brain Activity Classifiers Using Image-To-Image Transfer by Generative Adversarial Network.

Deep neural networks (DNNs) can accurately decode task-related information from brain activations. However, because of the non-linearity of DNNs, it is generally difficult to explain how and why they assign certain behavioral tasks to given brain activations, either correctly or incorrectly. One of the promising approaches for explaining such a black-box system is counterfactual explanation. In this framework, the behavior of a black-box system is explained by comparing real data and realistic synthetic data that are specifically generated such that the black-box system outputs an unreal outcome. The explanation of the system's decision can be explained by directly comparing the real and synthetic data. Recently, by taking advantage of advances in DNN-based image-to-image translation, several studies successfully applied counterfactual explanation to image domains. In principle, the same approach could be used in functional magnetic resonance imaging (fMRI) data. Because fMRI datasets often contain multiple classes (e.g., multiple behavioral tasks), the image-to-image transformation applicable to counterfactual explanation needs to learn mapping among multiple classes simultaneously. Recently, a new generative neural network (StarGAN) that enables image-to-image transformation among multiple classes has been developed. By adapting StarGAN with some modifications, here, we introduce a novel generative DNN (counterfactual activation generator, CAG) that can provide counterfactual explanations for DNN-based classifiers of brain activations. Importantly, CAG can simultaneously handle image transformation among all the seven classes in a publicly available fMRI dataset. Thus, CAG could provide a counterfactual explanation of DNN-based multiclass classifiers of brain activations. Furthermore, iterative applications of CAG were able to enhance and extract subtle spatial brain activity patterns that affected the classifier's decisions. Together, these results demonstrate that the counterfactual explanation based on image-to-image transformation would be a promising approach to understand and extend the current application of DNNs in fMRI analyses.