Self-Supervised Learning Application on COVID-19 Chest X-ray Image Classification Using Masked AutoEncoder.

The COVID-19 pandemic has underscored the urgent need for rapid and accurate diagnosis facilitated by artificial intelligence (AI), particularly in computer-aided diagnosis using medical imaging. However, this context presents two notable challenges: high diagnostic accuracy demand and limited availability of medical data for training AI models. To address these issues, we proposed the implementation of a Masked AutoEncoder (MAE), an innovative self-supervised learning approach, for classifying 2D Chest X-ray images. Our approach involved performing imaging reconstruction using a Vision Transformer (ViT) model as the feature encoder, paired with a custom-defined decoder. Additionally, we fine-tuned the pretrained ViT encoder using a labeled medical dataset, serving as the backbone. To evaluate our approach, we conducted a comparative analysis of three distinct training methods: training from scratch, transfer learning, and MAE-based training, all employing COVID-19 chest X-ray images. The results demonstrate that MAE-based training produces superior performance, achieving an accuracy of 0.985 and an AUC of 0.9957. We explored the mask ratio influence on MAE and found ratio = 0.4 shows the best performance. Furthermore, we illustrate that MAE exhibits remarkable efficiency when applied to labeled data, delivering comparable performance to utilizing only 30% of the original training dataset. Overall, our findings highlight the significant performance enhancement achieved by using MAE, particularly when working with limited datasets. This approach holds profound implications for future disease diagnosis, especially in scenarios where imaging information is scarce.

Ex Vivo Drug Screening Assay with Artificial Membranes: Characterizing Cholesterol Desorbing Competencies of Beta-Cyclodextrins.

Despite advancements in contemporary therapies, cardiovascular disease from atherosclerosis remains a leading cause of mortality worldwide. Supported lipid bilayers (SLBs) are membrane interfaces that can be constructed with varying lipid compositions. Herein, we use a solvent-assisted lipid bilayer (SALB) construction method to build SLB membranes with varying cholesterol compositions to create a lipid-sterol interface atop a piezoelectric sensor. These cholesterol-laden SLBs were utilized to investigate the mechanisms of various cholesterol-lowering drug molecules. Within a flow-cell, membranes with varying cholesterol content were exposed to cyclodextrins 2-hydroxypropyl-beta-cyclodextrin (HPßCD) and methyl-beta-cyclodextrin (MßCD). Quartz-crystal microgravimetry with dissipation monitoring (QCM-D) enabled the collection of in vitro, real-time changes in relative areal mass and dissipation. We define the cholesterol desorbing competency of a cyclodextrin species via measures of the rate of cholesterol removal, the rate of the transfer of membrane-bound cholesterol to drug-complexed cholesterol, and the binding strength of the drug to the cholesterol-ladened membrane. Desorption data revealed distinct cholesterol removal kinetics for each cyclodextrin while also supporting a model for the lipid-cholesterol-drug interface. We report that MßCD removes a quantity of cholesterol 1.61 times greater, with a speed 2.12 times greater, binding affinity to DOPC lipid interfaces 1.97 times greater, and rate of internal cholesterol transfer 3.41 times greater than HPßCD.

Long-term outcomes of relmacabtagene autoleucel in Chinese patients with relapsed/refractory large B-cell lymphoma: Updated results of the RELIANCE study.

BACKGROUND AIMS: The RELIANCE study has demonstrated the activity and safety of relmacabtagene autoleucel (relma-cel) (JW Therapeutics [Shanghai] Co, Ltd, Shanghai, China), a CD19-targeted chimeric antigen receptor T-cell product, in patients with heavily pre-treated relapsed/refractory large B-cell lymphoma (r/r LBCL). This study aimed to report the updated 2-year data of the RELIANCE study. METHODS: The RELIANCE study (NCT04089215) was an open-label, multi-center, randomized, phase 1/2 registrational clinical trial conducted at 10 clinical sites in China. Adult patients with heavily pre-treated r/r LBCL were enrolled and received lymphodepletion chemotherapy followed by infusion of 100 × 106 or 150 × 106 relma-cel. The primary endpoint was objective response rate (ORR) at 3 months, as assessed by investigators. Secondary endpoints were duration of response (DoR), progression-free survival (PFS), overall survival (OS) and safety profiles. RESULTS: From November 2017 to January 2022, a total of 68 patients were enrolled, and 59 patients received relma-cel infusion. As of March 29, 2022, a total of 59 patients had a median follow-up of 17.9 months (range, 0.3-25.6). ORR was 77.59% (95% confidence interval [CI], 64.73-87.49) and complete response rate was 53.45% (95% CI, 39.87-66.66). Median DoR was 20.3 months (95% CI, 4.86-not reached [NR]) and median PFS was 7.0 months (95% CI, 4.76-24.15). Median OS was NR and 1-year and 2-year OS rates were 75.0% and 69.3%, respectively. Three (5.1%) patients experienced grade ≥3 cytokine release syndrome and two (3.4%) patients had grade ≥3 neurotoxicity. CONCLUSIONS: The updated data of the RELIANCE study demonstrate durable response with and manageable safety profile of relma-cel in patients with heavily pre-treated r/r LBCL.

Efficient Training on Alzheimer's Disease Diagnosis with Learnable Weighted Pooling for 3D PET Brain Image Classification.

Three-dimensional convolutional neural networks (3D CNNs) have been widely applied to analyze Alzheimer's disease (AD) brain images for a better understanding of the disease progress or predicting the conversion from cognitively impaired (CU) or mild cognitive impairment status. It is well-known that training 3D-CNN is computationally expensive and with the potential of overfitting due to the small sample size available in the medical imaging field. Here we proposed a novel 3D-2D approach by converting a 3D brain image to a 2D fused image using a Learnable Weighted Pooling (LWP) method to improve efficient training and maintain comparable model performance. By the 3D-to-2D conversion, the proposed model can easily forward the fused 2D image through a pre-trained 2D model while achieving better performance over different 3D and 2D baselines. In the implementation, we chose to use ResNet34 for feature extraction as it outperformed other 2D CNN backbones. We further showed that the weights of the slices are location-dependent and the model performance relies on the 3D-to-2D fusion view, with the best outcomes from the coronal view. With the new approach, we were able to reduce 75% of the training time and increase the accuracy to 0.88, compared with conventional 3D CNNs, for classifying amyloid-beta PET imaging from the AD patients from the CU participants using the publicly available Alzheimer's Disease Neuroimaging Initiative dataset. The novel 3D-2D model may have profound implications for timely AD diagnosis in clinical settings in the future.

Bleeding and thrombotic events in bevacizumab-treated patients with colorectal cancer on novel oral anticoagulants and antiplatelet medications.

Background: Bevacizumab is a humanized monoclonal anti-VEGF antibody often given in combination with fluorouracil-based chemotherapy as therapy for metastatic colorectal cancer (mCRC). The bleeding and thrombotic event rates in the setting of concurrent novel oral anticoagulants with and without aspirin and bevacizumab treatment in patients with mCRC remain unclear. Methods: 462 patients with mCRC at Barnes-Jewish Hospital were identified between December 1, 2016 and December 1, 2021 and screened for concurrent treatment with bevacizumab and anticoagulant or antiplatelet therapy. Demographic and clinical information was extracted by electronic chart review. Results: 21 patients were identified who received bevacizumab and either apixaban or rivaroxaban for mCRC treatment. Aspirin was prescribed in some of these patients within three years of starting apixaban or rivaroxaban. Of the 13 patients without aspirin prescription, nine were given apixaban, and four were given rivaroxaban while on bevacizumab. Four out of nine of the patients who received apixaban had epistaxis, and only one case resulted in any treatment discontinuation. Three out of four of the patients who received rivaroxaban experienced bleeding, and one of these three patients discontinued bevacizumab. We also looked at eight patients who had received aspirin. Two out of seven patients who received apixaban/bevacizumab/aspirin experienced bleeding and discontinued a medication. The patient who received rivaroxaban/bevacizumab/aspirin experienced bleeding and discontinued bevacizumab. No patient experienced adverse thrombotic events. Conclusions: Patients with mCRC treated with bevacizumab and apixaban with no history of aspirin use within three years have a relatively low risk of bleeding that warrants treatment discontinuation.

Human ZBP1 induces cell death-independent inflammatory signaling via RIPK3 and RIPK1.

ZBP1 is an interferon-induced cytosolic nucleic acid sensor that facilitates antiviral responses via RIPK3. Although ZBP1-mediated programmed cell death is widely described, whether and how it promotes inflammatory signaling is unclear. Here, we report a ZBP1-induced inflammatory signaling pathway mediated by K63- and M1-linked ubiquitin chains, which depends on RIPK1 and RIPK3 as scaffolds independently of cell death. In human HT29 cells, ZBP1 associated with RIPK1 and RIPK3 as well as ubiquitin ligases cIAP1 and LUBAC. ZBP1-induced K63- and M1-linked ubiquitination of RIPK1 and ZBP1 to promote TAK1- and IKK-mediated inflammatory signaling and cytokine production. Inhibition of caspase activity suppressed ZBP1-induced cell death but enhanced cytokine production in a RIPK1- and RIPK3 kinase activity-dependent manner. Lastly, we provide evidence that ZBP1 signaling contributes to SARS-CoV-2-induced cytokine production. Taken together, we describe a ZBP1-RIPK3-RIPK1-mediated inflammatory signaling pathway relayed by the scaffolding role of RIPKs and regulated by caspases, which may induce inflammation when ZBP1 is activated below the threshold needed to trigger a cell death response.

Reformation of the chondroitin sulfate glycocalyx enables progression of AR-independent prostate cancer.

Lineage plasticity of prostate cancer is associated with resistance to androgen receptor (AR) pathway inhibition (ARPI) and supported by a reactive tumor microenvironment. Here we show that changes in chondroitin sulfate (CS), a major glycosaminoglycan component of the tumor cell glycocalyx and extracellular matrix, is AR-regulated and promotes the adaptive progression of castration-resistant prostate cancer (CRPC) after ARPI. AR directly represses transcription of the 4-O-sulfotransferase gene CHST11 under basal androgen conditions, maintaining steady-state CS in prostate adenocarcinomas. When AR signaling is inhibited by ARPI or lost during progression to non-AR-driven CRPC as a consequence of lineage plasticity, CHST11 expression is unleashed, leading to elevated 4-O-sulfated chondroitin levels. Inhibition of the tumor cell CS glycocalyx delays CRPC progression, and impairs growth and motility of prostate cancer after ARPI. Thus, a reactive CS glycocalyx supports adaptive survival and treatment resistance after ARPI, representing a therapeutic opportunity in patients with advanced prostate cancer.

Adjunctive antidepressants for the acute treatment of bipolar depression: A systematic review and meta-analysis.

BACKGROUND: The depressive phase of bipolar disorder causes significant functional impairment and disease burden. The efficacy and safety of antidepressants in the treatment of bipolar depression has long been a subject of debate. AIMS: To synthesize evidence of the effectiveness, risk of mood switching, and tolerability of adjunctive antidepressants in acute bipolar depression compared to using mood stabilizers or antipsychotics alone. METHOD: Multiple databases were searched for randomized controlled trials, including open label and double-blinded, for patients ages 18 or older with acute bipolar depression, comparing efficacy and adverse events in those who used adjunctive antidepressants versus without. Risk of bias and outcomes were assessed using the Cochrane Risk of Bias Tool. This study has PROSPERO registration CRD42016037701. RESULTS: Nineteen studies met inclusion criteria. Adjunctive antidepressants showed no significant effect on improving response rate (RR=1.10, 95%CI: 0.98-1.23). Subgroup analysis showed that adjunctive antidepressants with antipsychotics had a small but significantly better response rate compared to antipsychotics alone, which was not seen with adjunctive antidepressants with mood stabilizers. However, that finding was limited by studies predominantly using olanzapine as the antipsychotic medication. Adjunctive antidepressants had no clinically significant impact (but a small statistically significant impact) on improving depressive symptoms (SMD=-0.13, 95%CI: -0.24 to -0.02). There was no association with increased mood switch (RR=0.97, 95%CI: 0.68-1.39) and there was an association with lower dropout due to inefficacy (RR=0.66, 95%CI: 0.45∼0.98). CONCLUSIONS: There is no evidence of adjunctive antidepressants clinically improving response rate or depressive symptoms for acute bipolar depression. They are well tolerated, without increasing the risk of short-term mood switch.

Internalization and trafficking of CSPG-bound recombinant VAR2CSA lectins in cancer cells.

Proteoglycans are proteins that are modified with glycosaminoglycan chains. Chondroitin sulfate proteoglycans (CSPGs) are currently being exploited as targets for drug-delivery in various cancer indications, however basic knowledge on how CSPGs are internalized in tumor cells is lacking. In this study we took advantage of a recombinant CSPG-binding lectin VAR2CSA (rVAR2) to track internalization and cell fate of CSPGs in tumor cells. We found that rVAR2 is internalized into cancer cells via multiple internalization mechanisms after initial docking on cell surface CSPGs. Regardless of the internalization pathway used, CSPG-bound rVAR2 was trafficked to the early endosomes in an energy-dependent manner but not further transported to the lysosomal compartment. Instead, internalized CSPG-bound rVAR2 proteins were secreted with exosomes to the extracellular environment in a strictly chondroitin sulfate-dependent manner. In summary, our work describes the cell fate of rVAR2 proteins in tumor cells after initial binding to CSPGs, which can be further used to inform development of rVAR2-drug conjugates and other therapeutics targeting CSPGs.

Impact of the 2020 New Zealand COVID-19 lockdown on participants in a community-based, peer-led fall prevention program.

OBJECTIVES: To explore the impacts of the 2020 New Zealand COVID-19 lockdown on peer-led Steady as You Go (SAYGO) fall prevention exercise classes and members, and to develop recommendations for mitigating impacts during future lockdowns. METHODS: Semi-structured phone interviews were conducted with 20 SAYGO program participants and managers following the first COVID-19 lockdown in New Zealand. Interviews were audio-recorded, transcribed verbatim and analysed using the General Inductive Approach. RESULTS: Participants were between 67 and 88 years of age, predominantly female (90%) and NZ European (80%), with one participant identifying as NZ Maori. Three themes were constructed from the analysis: Personal Function and Well-Being, Class Functioning and Logistics, and Future Strategies for Classes During Prospective Lockdowns. Participants used a range of strategies to stay connected with each other and continue the SAYGO exercises at home. Most participants and peer-leaders reported that they maintained physical function during lockdown, although some had feelings of psychological distress and social isolation. Contact systems and resource distribution varied substantially between groups. Classes resumed post-lockdown with only minor modifications and slightly decreased attendance. CONCLUSIONS: Overall, members of this peer-led model of fall prevention classes demonstrated resilience during the COVID-19 lockdown, despite some challenges. We propose three recommendations to address the challenges of maintaining existing peer-led exercise classes in the context of prospective lockdowns: (1) develop a comprehensive contact detail register and plans for each group; (2) delivery of modified exercise classes remotely over lockdown; and (3) implementation of a nationwide IT education and resource program for older adults.

When are intergroup attitudes judged as free speech and when as prejudice? A social identity analysis of attitudes towards immigrants.

Although anti-immigrant attitudes continue to be expressed around the world, identifying these attitudes as prejudice, truth or free speech remains contested. This contestation occurs, in part, because of the absence of consensually agreed-upon understandings of what prejudice is. In this context, the current study sought to answer the question, "what do people understand to be prejudice?" Participants read an intergroup attitude expressed by a member of their own group (an "in-group" member) or another group (an "out-group" member). This was followed by an interpretation of the attitude as either "prejudiced" or "free speech." This interpretation was also made by in-group or an out-group member. Subsequent prejudice judgements were influenced only by the group membership of the person expressing the initial attitude: the in-group member's attitude was judged to be less prejudiced than the identical attitude expressed by an out-group member. Participants' judgements of free speech, however, were more complex: in-group attitudes were seen more as free speech than out-group attitudes, except when an in-group member interpreted those attitudes as prejudice. These data are consistent with the Social Identity Approach to intergroup relations, and have implications for the processes by which intergroup attitudes become legitimised as free speech instead of prejudice.

Overcoming glutamate auxotrophy in Escherichia coli itaconate overproducer by the Weimberg pathway.

Biosynthesis of itaconic acid occurs through decarboxylation of the TCA cycle intermediate cis-aconitate. Engineering of efficient itaconate producers often requires elimination of the highly active isocitrate dehydrogenase to conserve cis-aconitate, leading to 2-ketoglutarate auxotrophy in the producing strains. Supplementation of glutamate or complex protein hydrolysate then becomes necessary, often in large quantities, to support the high cell density desired during itaconate fermentation and adds to the production cost. Here, we present an alternative approach to overcome the glutamate auxotrophy in itaconate producers by synthetically introducing the Weimberg pathway from Burkholderia xenovorans for 2-ketoglutarate biosynthesis. Because of its independence from natural carbohydrate assimilation pathways in Escherichia coli, the Weimberg pathway is able to provide 2-ketoglutarate using xylose without compromising the carbon flux toward itaconate. With xylose concentration carefully tuned to minimize excess 2-ketoglutarate flux in the stationary phase, the final strain accumulated 20 g/L of itaconate in minimal medium from 18 g/L of xylose and 45 g/L of glycerol. Necessity of the recombinant Weimberg pathway for growth also allowed us to maintain multi-copy plasmids carrying in operon the itaconate-producing genes without addition of antibiotics. Use of the Weimberg pathway for growth restoration is applicable to other production systems with disrupted 2-ketoglutarate synthesis.

Reduction of Derlin activity suppresses Notch-dependent tumours in the C. elegans germ line.

Regulating the balance between self-renewal (proliferation) and differentiation is key to the long-term functioning of all stem cell pools. In the Caenorhabditis elegans germline, the primary signal controlling this balance is the conserved Notch signaling pathway. Gain-of-function mutations in the GLP-1/Notch receptor cause increased stem cell self-renewal, resulting in a tumour of proliferating germline stem cells. Notch gain-of-function mutations activate the receptor, even in the presence of little or no ligand, and have been associated with many human diseases, including cancers. We demonstrate that reduction in CUP-2 and DER-2 function, which are Derlin family proteins that function in endoplasmic reticulum-associated degradation (ERAD), suppresses the C. elegans germline over-proliferation phenotype associated with glp-1(gain-of-function) mutations. We further demonstrate that their reduction does not suppress other mutations that cause over-proliferation, suggesting that over-proliferation suppression due to loss of Derlin activity is specific to glp-1/Notch (gain-of-function) mutations. Reduction of CUP-2 Derlin activity reduces the expression of a read-out of GLP-1/Notch signaling, suggesting that the suppression of over-proliferation in Derlin loss-of-function mutants is due to a reduction in the activity of the mutated GLP-1/Notch(GF) receptor. Over-proliferation suppression in cup-2 mutants is only seen when the Unfolded Protein Response (UPR) is functioning properly, suggesting that the suppression, and reduction in GLP-1/Notch signaling levels, observed in Derlin mutants may be the result of activation of the UPR. Chemically inducing ER stress also suppress glp-1(gf) over-proliferation but not other mutations that cause over-proliferation. Therefore, ER stress and activation of the UPR may help correct for increased GLP-1/Notch signaling levels, and associated over-proliferation, in the C. elegans germline.

A Retrospective Examination of the Impact of Pharmacotherapy on Parent-Child Interaction Therapy.

Objective: Parent-child interaction therapy (PCIT) is an evidence-based approach for children aged 2-7 years with disruptive behavior problems. This study examined the effectiveness of PCIT with and without concurrent pharmacotherapy. Methods: A convenience sample was collected from a retrospective chart review of preschool-aged children treated with PCIT at the Mayo Clinic Young Child Clinic between 2016 and 2020. Quantitative and qualitative data were abstracted from all patients. The sample was divided into two groups based on psychotropic medications status (medicated and unmedicated) at the initiation of PCIT. Effectiveness of treatment was assessed with the change in Eyberg Child Behavior Inventory (ECBI) score. The change over time in ECBI score was compared between the two PCIT groups with and without concurrent pharmacotherapy using a linear mixed model. Results: Of the 62 youth, 38.71% were females. Mean age was 4.71 ± 1.17 years. The mean baseline ECBI score was 148.74 ± 30.86, indicating clinically significant disruptive behaviors. The mean number of PCIT sessions was 6.59 ± 3.82. There was no statistically significant difference in ECBI scores between the two groups at pre-PCIT (medication group: 149.68, standard error [SE] = 11.61 vs. unmedicated group: 147.92, SE = 10.93, p = 0.8904) and at post-PCIT (medication group: 116.27 [SE = 11.89] vs. unmedicated group: 128.86 [SE = 11.57], p = 0.3464). There was a statistically significant improvement in ECBI scores for both groups after completing therapy (medication group = -33.41 [-22.32%], SE = 6.27, p < 0.0001; d = 1.144; unmedicated group = -19.06 [-12.88%], SE = 5.78, p = 0.0022; d = 1.078). Conclusions: PCIT reduced disruptive behaviors in this sample of young children regardless of concurrent pharmacotherapy. Future prospective studies should consider one particular pharmacological agent and long-term outcomes of treatment. PCIT and certain pharmacological treatments could have complex and important bidirectional priming effects for both treatments.

Glycaemic control across the lifespan in a cohort of New Zealand patients with type 1 diabetes mellitus.

BACKGROUND: It is well known that tight glycaemic control reduces all-cause mortality and the development of microvascular complications in patients with type 1 diabetes mellitus (T1D), but that effective glycaemic control is difficult to achieve in different age groups. Currently, the state of glycaemic control across the lifespan in patients with T1D in New Zealand is not known. AIM: To determine the differences in glycaemic control with age, gender, rurality and ethnicity in patients with T1D in the Waikato region of New Zealand. METHODS: Retrospective review of clinical records of all patients with T1D on the Waikato Regional Diabetes Database in December 2017 (n = 1303). Glycaemic control was determined by the most recent HbA1c in the past 2 years. RESULTS: Median (25%, 75%) HbA1c was 67 (59, 81) mmol/mol (8.3%) and highest in those aged 15-29 years. Values exceeded clinical recommendations in 85.3% of all patients. Median HbA1c was lower in patients on insulin pump therapy than on multiple daily injections (63 (7.9%) versus 69 mmol/mol (8.5%); P < 0.001), though insulin pumps were significantly less likely to be used by Maori (P = 0.003) and men (P < 0.0001). Worsening glycaemic control was associated with increasing social deprivation (P < 0.001) but was not influenced by rural/urban living. CONCLUSIONS: Poor glycaemic control in Waikato patients with T1D is likely due to inequities in health care, including reduced access to insulin pump therapy, particularly in Maori and socially deprived populations.

ß-amyloid and tau drive early Alzheimer's disease decline while glucose hypometabolism drives late decline.

Clinical trials focusing on therapeutic candidates that modify ß-amyloid (Aß) have repeatedly failed to treat Alzheimer's disease (AD), suggesting that Aß may not be the optimal target for treating AD. The evaluation of Aß, tau, and neurodegenerative (A/T/N) biomarkers has been proposed for classifying AD. However, it remains unclear whether disturbances in each arm of the A/T/N framework contribute equally throughout the progression of AD. Here, using the random forest machine learning method to analyze participants in the Alzheimer's Disease Neuroimaging Initiative dataset, we show that A/T/N biomarkers show varying importance in predicting AD development, with elevated biomarkers of Aß and tau better predicting early dementia status, and biomarkers of neurodegeneration, especially glucose hypometabolism, better predicting later dementia status. Our results suggest that AD treatments may also need to be disease stage-oriented with Aß and tau as targets in early AD and glucose metabolism as a target in later AD.

A Randomised Control Trial and Comparative Analysis of Multi-Dimensional Learning Tools in Anatomy.

This article presents the results of a study that examined students' ability to retain what they have learned in an anatomy course after thirty days via using various learning tools for twenty minutes. Fifty-two second-year medical students were randomly assigned to three learning tools: text-only, three-dimension visualisation in a two-dimensional screen (3DM), or mixed reality (MR). An anatomy test lasting for twenty minutes measuring spatial and nominal knowledge was taken immediately after the learning intervention and another thirty days later. Psychometric tests were also used to measure participants' memory, reasoning and concentration abilities. Additionally, electroencephalogram data was captured to measure the participants' awakeness during the learning session. Results of this study showed that the MR group performed poorly in the nominal questions compared to the other groups; however, the MR group demonstrated higher retention in both the nominal and spatial type information for at least a month compared to the other groups. Furthermore, participants in the 3DM and MR groups reported increased engagement. The results of this study suggest that three-dimensional visualiser tools are likely to enhance learning in anatomy education. However, the study itself has several limitations; some include limited sample size and various threats to internal validity.

Dynamic Image for 3D MRI Image Alzheimer's Disease Classification.

We propose to apply a 2D CNN architecture to 3D MRI image Alzheimer's disease classification. Training a 3D convolutional neural network (CNN) is time-consuming and computationally expensive. We make use of approximate rank pooling to transform the 3D MRI image volume into a 2D image to use as input to a 2D CNN. We show our proposed CNN model achieves 9.5% better Alzheimer's disease classification accuracy than the baseline 3D models. We also show that our method allows for efficient training, requiring only 20% of the training time compared to 3D CNN models. The code is available online: https://github.com/UkyVision/alzheimer-project.