Machine Learning based Point-of-Care Disease Diagnostics using Dried patterns formed by E. coli bacteria-laden Sessile Urine Droplets

Machine Learning-based Morphological Pattern Analysis of Dried Deposits formed by E. coli-laden Sessile Urine Droplets for Point-of-Care Diagnostics | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Machine Learning-based Morphological Pattern Analysis of Dried Deposits formed by E. coli -laden Sessile Urine Droplets for Point-of-Care Diagnostics View ORCID Profile M Ashwin Ganesh , Sophia M , Jason Joy Poopady , Abdur Rasheed , Durbar Roy , Amey Nitin Agharkar , Visakh Vaikuntanathan , Kirti Parmar , Dipshikha Chakravortty , Saptarshi Basu doi: https://doi.org/10.1101/2025.10.14.25337322 M Ashwin Ganesh a Department of Mechanical Engineering, Indian Institute of Science , Bengaluru, Karnataka, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for M Ashwin Ganesh Sophia M a Department of Mechanical Engineering, Indian Institute of Science , Bengaluru, Karnataka, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jason Joy Poopady a Department of Mechanical Engineering, Indian Institute of Science , Bengaluru, Karnataka, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site Abdur Rasheed b Aix Marseille University, CNRS, IUSTI , Marseille, France Find this author on Google Scholar Find this author on PubMed Search for this author on this site Durbar Roy c International Center for Theoretical Sciences, Tata Institute of Fundamental Research , Bengaluru, Karnataka, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site Amey Nitin Agharkar a Department of Mechanical Engineering, Indian Institute of Science , Bengaluru, Karnataka, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site Visakh Vaikuntanathan d Department of Mechanical Engineering, Shiv Nadar Institution of Eminence , Greater Noida, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kirti Parmar e Division of Biological Sciences, Department of Microbiology and Cell Biology, Indian Institute of Science , Bengaluru, Karnataka, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site Dipshikha Chakravortty e Division of Biological Sciences, Department of Microbiology and Cell Biology, Indian Institute of Science , Bengaluru, Karnataka, India f Adjunct Faculty, School of Biology, Indian Institute of Science Education and Research , Thiruvananthapuram, Kerala, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: dipa{at}iisc.ac.in Saptarshi Basu a Department of Mechanical Engineering, Indian Institute of Science , Bengaluru, Karnataka, India Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: sbasu{at}iisc.ac.in Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Urinary Tract Infections (UTIs) are a rising global health concern, primarily caused by E. coli , disproportionately affecting women and the elderly. Despite recent advancements in diagnostic techniques, a critical gap persists: time delays, high costs, and false-positive predictions. This drives the requirement for rapid, reliable, and low-cost point-of-care diagnostic tools. As a vital step towards addressing this need, we present a machine learning-based morphological pattern analysis of dried deposits formed by E. coli-laden sessile urine droplets. Urine samples inoculated with E. coli at three distinct concentrations were dried and imaged using brightfield microscopy for pattern analysis. The objective of this study is twofold. First, we develop a supervised ternary classification model that identifies dried deposit images by bacterial concentration. The model employs multiple convolutional backbones for complementary feature extraction. It further refines these representations with an attentive fusion mechanism and predicts outcomes with a final stacking ensemble layer. Second, we aim to estimate a qualitative severity factor by applying cluster validity metrics to image-based features extracted from the trained classification model. In addition, we conducted Scanning Electron Microscopy (SEM) to gain physical insights into pattern formation associated with bacterial presence. Overall, this study introduces a robust proof-of-concept methodology for analyzing E. coli-laden dried deposits. It offers a rapid, low-cost, and scalable foundation with strong potential for integration into cyber-physical systems and real-time point-of-care diagnostic tools in resource-limited settings. Download figure Open in new tab Highlights Controlled evaporation experiments were conducted using urine samples inoculated with E. coli at three distinct concentrations, and the final dried deposits subsequently imaged using brightfield microscopy were utilized for morphological pattern analysis. Implemented machine learning-based morphological pattern classification that fuses diverse feature representations, further refines them using an attentive-feature fusion mechanism before the final prediction with a stacking ensemble layer. Achieved robust predictive performance with an overall accuracy of about 86%. Furthermore, the learned high-dimensional feature embeddings were projected onto lower-dimensional spaces using PCA and t-SNE plots. Developed a novel Qualitative Severity metric employing cluster validity indices (CVIs) to the extracted image-based discriminative features to provide an insight into deviation in pattern formation due to bacterial presence. In addition, Scanning Electron Microscopy (SEM) was conducted to gain physical insights into the impact of bacterial presence on the final pattern. Introduction Urinary Tract Infections (UTIs) are one of the most common bacterial infections worldwide, with more than 150 million cases being reported annually. Escherichia coli ( E. coli ) bacteria are the primary causal agent of this infection [ 1 ]. It is significantly more common among adult women, and the risk is almost doubled beyond the age of 65 [ 2 ]. The most common diagnostic techniques include the symptoms themselves, urinalysis, and the traditional culture-based technique (referred to as the gold standard). However, these methods have notable shortcomings, including time consumption and high cost, among others [ 3 ]. Precise official estimates for the share of UTI cases arising in rural/remote populations remain scarce, but multiple global assessments highlight a persistent diagnostic gap at the primary-care level in low-resource settings: limited laboratory capacity, long turnaround times for urine culture, out-of-pocket costs, and distance to facilities; conditions typical of rural areas that collectively lead to under-diagnosis [ 4 ]. Several modern novel diagnosis techniques, such as molecular diagnostics, Point of Care testing (POCT) methods, and artificial intelligence-based techniques, have been emerging in order to overcome the above drawbacks [ 5 ]. Collectively, these factors highlight the crucial need for developing rapid, low-cost, accessible, and reliable point-of-care UTI diagnosis. Convolutional Neural network (CNN) and Transformer based architectures are the dominant methods for medical image classification, which consistently outperform traditional machine learning techniques in this aspect and have led to several breakthroughs [ 6 , 7 ]. ResNet (Residual Neural Network) [ 8 ] mitigated the degradation and vanishing gradient problems using skip connections, which use identity shortcuts for gradient flow, preventing information loss. EfficientNet [ 9 ] introduced Compound Scaling (synchronized expansion of a network’s depth, width, and resolution to maximize accuracy) to overcome inefficient manual scaling problems in order to achieve high performance and exceptional accuracy. Inception [ 10 ] involved judicious kernel size selection, which enables processing images through multiple filter sizes simultaneously to capture fine details and broad patterns within a single layer. Combining such model architectures with transfer learning has proven powerful, especially while handling relatively limited and complex medical image datasets [ 11 ]. Vision transformers (ViT) [ 12 ] redefined the state-of-the-art by leveraging global self-attention, a mechanism that enables establishing broad contextual relationships. Following this, several novel architectures such as Swin [ 13 ], MaxViT [ 14 ], ConvNeXt [ 15 ] were introduced for enhanced performance using modernized design principles. Further, hybrid models like Inception-ResNet [ 16 ], CoAtNet [ 17 ], and MobileViT [ 18 ] have demonstrated exceptional performance by synergizing the capabilities from multiple architectures. However, CNNs were chosen over Transformers for this study because of their strong spatial inductive biases. Unlike Transformers, CNNs efficiently preserve fine-grained local features with lower computational complexity, making them particularly effective for our limited training dataset [ 19 ]. Artificial intelligence–driven medical image analysis has been undergoing tremendous advancement and widespread adoption, with evolving models capable of processing enormously high volumes of image data and recognizing complex patterns that were earlier undetectable [ 20 ]. A few among the prominent works based on medical image classification have been discussed here. Zahoor et al. [ 21 ] performed Brain Tumor classification using Magnetic Resonance Imaging (MRI) images based on their proposed Res-BRNet, integrating residual and spatial blocks systematically for improving discriminative capacity and generalization. Esteva et al. performed dermatologist-level classification of skin cancer [ 22 ], demonstrating that an ImageNet-pretrained Google-Inception-v3 CNN architecture trained on a massive dataset of clinical images of skin lesions encompassing over 2,000 different diseases could classify skin cancer with a level of accuracy comparable to board-certified dermatologists. Azizi et al. [ 23 ] proposed a self-supervised learning-based training on a vast, smaller, unlabeled medical dataset. Apart from classification, which are proven techniques for diagnosis, medical image segmentation provides pixel-level clarity, facilitating precise, clinically relevant understanding, such as U-Net [ 24 ], which uses skip-connections to preserve precise local details. Hybrid models like TransUNet [ 25 ] add transformers to capture the long-range global context of the image. To further address the high-resolution requirements of medical imaging, hierarchical architectures such as Swin-Unet [ 26 ] facilitate efficient modeling of both local and global features. Overall, these rapid architectural advancements highlight the important role of artificial intelligence in transforming medical imaging by automating diagnosis and supporting robust clinical decision-making. Attentive Feature fusion aggregates features from multiple architectures and assigns dynamic weights through attention-based training [ 27 ]. Selective Kernel Method (SKNet) [ 28 ] enables adaptive feature selection to prioritize the most optimal representations. Squeeze and Excitation Network (SENet) [ 29 ], utilizes channel attention to evaluate dynamic weights for global average-pooled features using dimensionality reduction and non-linearity. Convolutional Block Attention Module (CBAM) [ 30 ] enhances this by performing both channel and spatial-attention based feature weighting. Effective Channel-Attention Network (ECA-Net) [ 31 ] introduces a convolution-based architecture, effectively overcoming the shortcomings of earlier Multilayer Perceptron (MLP) based attention mechanisms. Attentional Feature Fusion (AFF) [ 32 ] unifies these concepts using an iterative framework to integrate channel and spatial attention for robust multi-scale fusion. While feature fusion spans various fields across domains, a few notable implementations in the context of medical imaging are as follows. Rajaraman et al. [ 33 ] proposed an ensemble framework for automated detection of pediatric pneumonia in chest X-rays that concatenates features from multiple convolution-based model backbones, demonstrating enhanced performance. Unlike simple concatenation, which can combine features from multiple architectures, attentive feature fusion mechanisms are placed a step ahead, dynamically weighing the combined features, thereby potentially suppressing background features and prioritizing the most prominent ones. Chen et al. [ 34 ] introduced a multimodal framework for cancer diagnosis and prognosis that integrates histopathology and genomic features using a gated attention mechanism. Qezelbash-Chamak et al. [ 35 ] proposed ConvTransGFusion, a novel framework integrating features based on ConvNeXt and Swin Transformer backbones using a learnable dual-attention gating mechanism. Rahman et al. [ 36 ], in their hybrid framework ViT-CNN for histopathological image classification, employed concatenation-based fusion of Vision Transformer and CNN backbones. Shareef et al. [ 37 ] proposed an Adaptive Fusion Attention (AFA) block, which blends channel and spatial attention, especially for medical imaging applications. Although the potential of attentive feature fusion-based strategies has been well-established, their application in the context of droplet evaporation-based point-of-care diagnostics remains relatively underexplored. Our work specifically explores this mechanism by incorporating the Effective Channel-Attention Network (ECA-Net) applied on the fused feature vectors obtained from multiple convolutional model backbones for morphological pattern analysis of dried patterns formed by E. coli -laden sessile urine droplets. Droplet evaporation-based diagnostics operates by analyzing micro- to nanoliter bio-fluid droplets to infer disease-relevant information from their evaporation residues and dried morphologies [ 38 , 39 ]. Pathological conditions directly influence the composition of complex bio-colloidal fluids, including blood, saliva, tears, and urine, through alterations in the concentrations and interactions of macromolecules, electrolytes, and cellular constituents [ 40 ]. Foundational studies and comprehensive reviews in colloid science establish the diagnostic potential of evaporating biological fluids, focusing primarily on complex bio-colloids. These works detail how internal fluid motions, specifically Capillary and Marangoni flows, interact with gelation and crystallization to shape the final dried pattern. Furthermore, they define the standard optical measurement techniques involved, including bright-field and polarized imaging, as well as profilometry, required to systematically quantify these distinct morphological signatures [ 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 ]. Roy et al. [ 49 ] elucidated the evaporation physics of bacteria-laden blood droplets, using theoretically validated experiments to demonstrate how bacterial presence modifies internal flow, evaporation dynamics, and final deposition patterns. Pal. A et al. [ 50 ] examined the temporal evaporation dynamics using time-lapse microscopy, which provided evidence that dynamics carry additional diagnostic signal. Hegde et al. [ 51 ] utilized non-contact vapor modulation to alter the internal fluid dynamics of evaporating respiratory droplets, demonstrating that this controlled flow strictly dictates localized bacterial aggregation and the subsequent deposit pattern formation. Urine, being rich in electrolytes, proteins, and metabolites, produces characteristic dried patterns (concentric rings, crystal fans, dendrites, crack networks) whose features correlate with composition and pathology [ 52 ]. The application of machine learning and pattern recognition techniques to analyze dried deposit formations is a rapidly emerging field, driven by several contemporary studies [ 53 ]. O. Hegde et al. [ 54 ] designed a convolution-based architecture as a Point-of-Care diagnostic tool to classify the dried blood droplet images into seven bacterial infections. Demir et al. [ 55 ] studied whole blood and urine droplet samples to detect bladder cancer using pretrained ResNet-based architectures. Yang et al. [ 52 ] performed label-free urinary protein quantification employing an Inception-ResNet architecture, to serve as a promising alternative to traditional techniques involved. Building upon these foundational studies, our present work investigates the variation in residual morphological patterns obtained from E. coli -laden sessile urine droplets at various bacterial concentrations, employing a deep learning-based pattern recognition framework. Machine learning for UTI Diagnostics has showcased great potential by starting to serve as a rapid tool. As several researchers have proven through feeding multiple types of input data into deep learning models for making predictions [ 56 ] and demonstrating state-of-the-art performance. For predicting bacteriuria in the emergency department, Sheele et al. [ 57 ] utilized patient data as feature variables into an extreme gradient boosting (XGBoost) classifier, which predicted the infection with a reasonable accuracy. Farashi et al. [ 58 ] used demographic, urine test, and blood test data of patients as features into classifiers, including XGBoost, Light Gradient Boosting Machines (LightGBM), and Random Forest for prediction. A desktop application built by Arches et al. [ 59 ] used an image of a urine test strip captured by a smartphone as input, utilizing a CNN for extracting features from colored pads into numerical analyte levels. These features were further fed into the Support Vector Machine (SVM) for the final classification. Liou et al. [ 60 ] created a publicly available dataset consisting of microscopic images of liquid urine samples from symptomatic patients. A Patch U-Net model was trained to classify and give the count of granular elements, including white blood cells, red blood cells, and bacteria within them, as a precursor for further diagnosis. Large-volume microscopy-based study by Iriya et al. [ 61 ] captured the phenotypic features of bacterial motion, distinguishing them from other non-motile particles, wherein a deep neural network was trained based on numerical features summarizing the bacterial movement. Thereby, such contemporary studies collectively signify the strong potential of machine learning in diagnosing urinary tract infection, by leveraging diverse data modalities ranging from structured clinical data and test strip images to the direct microscopic analysis of bacteria and their movement. Severity estimation of urinary tract infection, being inherently complex, depends on a multitude of factors ranging from clinical signs to laboratory data. Clayson et al. [ 62 ] evaluated a severity metric by creating and validating the UTI Symptom Assessment (UTISA) questionnaire, to plot seven prominent features on two metrics, namely severity and bothersomeness, measured the severity on a 4-point scale, purely based on symptoms. A severity measure based on the presence of seven clinical features, by Hay et al. [ 63 ], had been estimated for UTI diagnosis in children. Another probabilistic approach from a diagnostic laboratory perspective by De Rosa et al. [ 64 ], combining results from standard urinalysis and automated flow cytometry into a single, predictive value called UTI-Risk Score, was evaluated using measured counts of certain specific constituents in the urine samples. Inferences based on these studies highlight the multifaceted nature of UTI severity, quantified through distinct lenses: patient-reported outcomes, clinical risk stratification, and objective laboratory diagnostics. Building upon this objective, we model diagnostic severity as a function of geometric feature deviations from healthy, baseline dried droplet morphology. We propose that bacterial concentration directly correlates with the degree of morphological pattern disruption within dried urine deposits. For this part of the analysis, we employ cluster validity indices (CVIs), which are statistical hypothesis tests [ 65 ], which help to evaluate the quality of structure obtained from clustering by comparing the inter and intra-cluster compactness and translating them into a numerical score. In this study, we utilized a subset of internal cluster validity indices which operate in an unsupervised manner, without prior knowledge of the class labels within a given cluster, namely Silhouette score [ 66 ], Davies-Bouldin Index [ 67 ], Calinski-Harabasz Index [ 68 ], and Centroidal distance which demonstrate consistent robustness across different types of data [ 69 ]. In this study, microscopic images of dried patterns formed by E. coli -laden sessile urine droplets at various bacterial concentrations were utilized for machine-learning-driven, morphological pattern classification and qualitative severity estimation. Controlled evaporation experiments were conducted using urine samples inoculated with E. coli at three distinct concentrations, and sub-sequently the dried deposits were imaged using brightfield microscopy. Furthermore, Scanning Electron Microscopy (SEM) was conducted to gain physical insights into the impact of bacterial presence on the final pattern. Our study included the dried patterns corresponding to the raw urine samples (without E. coli ) and the ones inoculated with E. coli at 10 9 CFU/mL, and 10 11 CFU/mL, hereafter referred to as U-EC-00, U-E-09 and U-EC-11, respectively. First, a supervised morphological pattern classification was performed, wherein dried deposit patterns were identified by bacterial concentration. In addition, the dried deposit formation behavior was analyzed visually using linear and nonlinear dimensionality-reduction maps. Our model uses an attentive feature fusion mechanism, namely the Effective Channel-Attention Network (ECA-Net) [ 31 ], which integrates the diverse feature representations from three deep convolutional architectures to assign attention-based weights to each feature dimension. The subsequent part of the analysis examined the dried deposits with changes in E. coli concentration, for a specific subject under consideration, by estimating a qualitative diagnostic severity factor based on this pattern variation, employing cluster validity indices (CVIs), applied to image-based features extracted from a trained pattern classification model, as an attempt to obtain a robust qualitative metric for diagnostic support. The major contributions of this work can be summarized as follows: (i) Implementing a deep learning model architecture involving an Attentive Feature Fusion-based mechanism using convolution-based architectural backbones as feature extractors, in the context of droplet evaporation-based morphological pattern analysis for point-of-care diagnostics. (ii) Developed a novel qualitative diagnostic severity factor, which leverages cluster validity metrics applied to the image-based discriminative feature clusters extracted from the trained classification model. Methodology Experiments and Data Acquisition Urine samples were obtained from 13 subjects, with appropriate institutional consent. The experimental procedure involved repeated sampling performed over time. Subjects were instructed to collect their first morning urine sample in two distinct sterile containers. One container was allocated to the pathological examination (chemical and microscopic analysis), while the other was reserved for the droplet evaporation studies, for which the raw urine was transferred to microcentrifuge tubes to inoculate with specific bacterial concentrations simulating Urinary Tract Infection (UTI) conditions. 500 µL of raw urine was taken in one microcentrifuge tube. The 10 9 CFU/mL concentration was prepared using 495 µL of raw urine with 5 µL of bacterial solution, and the 10 11 CFU/mL concentration using 450 µL of raw urine with 50 µL of bacterial solution. The evaporation experiments were conducted under controlled environmental conditions, maintaining a Relative Humidity (RH) of 45 ± 2% and ambient temperature of 28 ± 2 ◦ C. A 1.5 µ L droplet of the prepared sample was pipetted onto a coverslip. Evaporation of the sample was monitored using top-view optical microscopy. The resulting final patterns were captured using bright-field microscopy at 10× magnification. Fig. 1 illustrates the typical temporal variation of the normalized volume of the drying sessile urine droplet, demonstrating an increase in the drying time with the bacterial presence (at 10 11 CFU/mL) as compared to the scenario without bacterial presence. The figure also presents the pattern evolution imaged during the evaporation process. Download figure Open in new tab Figure 1: Temporal evolution of the normalized droplet volume during evaporation. Insets illustrate the sequential pattern formation comparing samples without E. coli (insets A1-A3) compared to samples with E. coli at 10 11 CFU/mL (insets B1-B3) In addition, scanning electron microscopy (SEM; JEOL-SEM IT 300 Microscopy facility) was employed to characterize the sub-micron architecture of the dried droplet residues at three E.coli concentrations (U-EC-00, U-EC-09, and U-EC-11). Sessile droplets were deposited and dried on standard glass slides (Blue Star, 75 mm × 25 mm). For imaging, the samples were mounted on aluminium stubs using double-sided carbon tape and were gold-coated. Backscattered electron (BSE) imaging was conducted in high-vacuum mode with an accelerating voltage of 15 kV, a working distance of approximately 12 mm, and a probe current of 50 nA. Fig. 2 illustrates the morphological details of the patterns, capturing the complete deposit as well as the center and edge sections. Download figure Open in new tab Figure 2: Representative images from Scanning Electron Microscopy of Dried patterns formed at various E. coli concentrations. Panel (1) shows the pattern without E. coli (U-EC-00), panel (2) with E. coli at 10 9 CFU/mL (U-EC-11), and panel (3) with E. coli at 10 11 CFU/mL (U-EC-11), each capturing the (A) complete deposit, (B) the central region, and the (C) edge portion of the deposit. Data Processing The microscopic images were organized as experimental subsets, where each subset consisted of dried patterns gathered from multiple runs (corresponding to the urine sample from a specific healthy individual on a specific day). Each subset consisted of three classes (distinct bacterial concentrations), namely U-EC-00, U-EC-09, and U-EC-11, respectively. Multiple sessile droplet evaporation experimental runs were performed from a given urine sample, with an average of 50 images per class within each subset. This yielded a total of 3060 images, distributed as 1027 images for U-EC-00, 1038 images for U-EC-09, and 995 images for U-EC-11, respectively, each of which was stored in RGB format with a high resolution of 6000 × 4000 pixels. Our data structuring, along with typical dried patterns observed from three chosen subsets, has been clearly depicted in Fig. 3 . To observe the model’s ability to generalize across different experimental conditions, the data were partitioned (for training and test) at a subset level. Of the 18 available subsets, 15 were allocated for training and validation, while the remaining 3 were reserved as a strictly held-out test set. This partitioning strategy ensured that the test data represented entirely unseen experimental contexts, providing a rigorous benchmark for the system’s robustness. Download figure Open in new tab Figure 3: The top panel illustrates a schematic overview of the dataset structure, organized as subsets, each comprising multiple runs at various E. coli concentrations. The bottom panel shows representative images of the typical final patterns from three chosen subsets. Subsequently, an image processing pipeline was designed to balance computational efficiency with the preservation of critical spatial features. Both the training set and a held-out test set were processed through the pipeline, where a five-fold stratified cross-validation scheme was implemented to partition the data. This ensured rigorous isolation between training and validation contexts while maintaining consistent class distributions across each fold. To focus on the primary regions of interest, firstly, the images were subjected to a 2500-pixel center crop before being resized to the target resolution of (1250x1250) using bicubic interpolation with antialiasing, to preserve fine spatial details and avoid distortion during resizing. Given the relatively limited sized dataset, the training folds were subjected to a comprehensive augmentation procedure to improve model robustness; this included random horizontal and vertical flips, rotations up to 180°, and 10% jitter in the color space (brightness, contrast, and saturation). Finally, all images across every subset were normalized using standard ImageNet statistics to align with the pretrained back-bone parameters. Specifically, each image X was normalized channel-wise as X ′ = ( X − µ ) /σ , where µ = (0.485, 0.456, 0.406) and σ = (0.229, 0.224, 0.225), corresponding to the RGB channels. The resulting data were loaded using PyTorch DataLoaders with a fixed batch size of 16 and num_workers = 16, ensuring efficient and consistent GPU utilization. Machine Learning Methodology The schematic of our two-fold objective and its entire workflow has been clearly elucidated in the Fig. 4 . The first objective of this study was to perform ternary classification of input images based on the bacterial concentrations (U-EC-00, U-EC-09, and U-EC-11). Following preprocessing, three ImageNet-pretrained convolutional architectures: ResNet-34, EfficientNet-B2, and DenseNet-121 were utilized as model backbones, fine-tuned to adapt the learned hierarchical features to the specific morphological markers of the dried patterns. These models were trained for a maximum of 50 epochs using the Adam optimizer and categorical cross-entropy loss, with the learning rate regulated by a cosine annealing scheduler to dynamically decay the rate from 10 −4 to 10 −6 for stable convergence near the global minima. An early stopping mechanism with a patience of 7 epochs was implemented to halt training and save the optimal model weights based on the minimum validation loss. Subsequently, the feature embeddings were extracted from the final convolutional layers of the three backbones using global average pooling. These individual feature vectors with dimensions of 512 for ResNet-34, 1408 for EfficientNet-B2, and 1024 for DenseNet121 were then concatenated into a single, unified, high-dimensional feature vector totaling 2944 dimensions. This composite feature space was refined using an attentive feature fusion block, a mechanism designed to dynamically re-weight feature channels based on their relative discriminatory importance. Specifically, a Effective Channel attention (ECA-Net) was employed to refine these representations by adaptively reweighting channel-wise features through a 1D convolution, enabling efficient modeling of local cross-channel interactions without introducing dimensionality reduction. Download figure Open in new tab Figure 4: Schematic describing the workflow of our machine learning pipeline for morphological pattern classification and qualitative severity estimation. For the final pattern classification, these attention-weighted fused feature vectors were fed into a stacking ensemble consisting of five base classifiers: Support Vector Machine [ 70 ], Random Forest [ 71 ], Naive Bayes [ 72 ], K-Nearest Neighbors [ 73 ], and Logistic Regression [ 74 ]. To ensure unbiased objective function convergence and optimal distributional scaling, features were Z-score normalized for distance-dependent and probabilistic models (SVM, KNN, Logistic Regression, and Naive Bayes), while the inherently scale-invariant Random Forest was trained on raw feature magnitudes. Each base model underwent exhaustive hyperparameter tuning using GridSearchCV paired with 5-fold cross-validation, an optimization procedure that systematically evaluates parameter combinations to identify the most robust configuration. To strictly prevent data leakage, out-of-fold probabilities from the base models were concatenated into a meta-feature space. A Multi-Layer Perceptron (MLP) meta-learner then processed these out-of-fold probabilities to output the final classification result. This stacking ensemble effectively enabled the synthesis of diverse strengths of the base classifiers to enhance final predictive robustness. To mitigate the severe risk of overfitting inherent to relying on a single feature extractor, our model architecture involved a hierarchical panel of experts approach, designed to capture diverse and complementary sets of morphological features. ResNet-34, EfficientNet-B2, and DenseNet-121 were specifically chosen as backbone architectures to synergistically leverage their respective strengths: residual learning, optimized compound scaling, and dense feature reuse, ensuring a di-verse and complementary spatial feature extraction. The resulting fused feature representations were subsequently refined through attentive feature fusion, a mechanism designed to amplify the more crucial discriminators and to suppress the relatively irrelevant ones. Finally, the architecture employs a stacking ensemble head, which employs a panel of classifiers, with the meta-learner designed to effectively integrate the heterogeneous predictions across diverse decision strategies to improve robustness. At a foundational level, CNNs were selected over Transformers because of their inherent inductive biases, specifically locality and translation invariance, which promote robust performance on limited-sized datasets without necessitating massive training data. Consequently, CNNs reliably extract and preserve the fine-grained spatial features critical for morphological classification. To visualize the high-dimensional feature vectors, a non-linear dimensionality reduction technique, t-distributed Stochastic Neighbor Embedding (t-SNE), and a linear map, Principal Component Analysis (PCA), were employed. PCA [ 75 ] provides a linear mapping of feature vectors onto a lower-dimensional subspace, defined by the directions of maximum variance; the projection was learned from the training data and subsequently applied to the test data to obtain a consistent representation. t-SNE [ 76 ] preserves local data topology by modeling the pairwise similarities in high dimensions using Gaussian kernels and in low dimensions using a heavy-tailed Student-t distribution, and computes a low-dimensional embedding by minimizing the KL divergence between these two distributions over all point pairs. The attention-weighted feature vectors were L 2 -normalized before projection to ensure a uniform scale, preventing high-magnitude components from disproportionately dominating the distance metrics and variance calculations inherent to these algorithms. The predictive capacity of the model on unseen data was evaluated through standard classification metrics, defined based on true positives ( TP k ), false positives ( FP k ), and false negatives ( FN k ) for each class k . Precision indicates how many of the samples predicted as class k are actually correct, while recall indicates how many of the true class k samples are correctly identified. The F1-score combines precision and recall through their harmonic mean to provide a single balanced measure. To assess performance across all classes without being affected by class imbalance, we report the macro-average, computed as the simple average of the metrics over all classes. In addition, overall accuracy is calculated as the fraction of correctly predicted samples out of the total number of test samples. Further, Receiver Operating Characteristic (ROC) curves were used, which characterize the True Positive Rate versus the False Positive Rate across varying decision thresholds. In addition, class-wise continuous Gaussian kernel density estimates were plotted for the magnitude of the attention-weighted feature vectors. The second objective was to estimate a diagnostic severity factor. Using the trained pattern classification model, we evaluated each subset from the held-out test set. The images from the given subset underwent standard preprocessing, feature extraction, and attentive feature fusion. Subsequently, a cluster-based analysis was conducted, evaluating four distinct cluster validity indices, namely the Silhouette Score, Calinski-Harabasz Index, Centroidal Distance, and Davies-Bouldin Index on the resulting attention-weighted fused features. This process yielded a normalized, qualitative severity metric ( S ∈ [0, 1]) representing subject-specific pathology. The cluster validity indices utilized for this part of the analysis have been clearly explained in Table 1 . View this table: View inline View popup Download powerpoint Table 1: The Chart of internal cluster validity indices Hypothesis for Qualitative Severity Estimation To establish a quantifiable measure of severity, we performed a proof-of-concept analysis to estimate a qualitative severity factor based on the discriminative features of dried deposit patterns within a given subset from the unseen test fold. Our approach is based upon the understanding that urine serves as a highly dynamic biochemical fingerprint [ 77 ]. Given that the chemical concentration in urine fluctuates significantly and is influenced by multiple systemic factors such as age, diet, health status, and medication, the resulting dried deposits, and consequently their spatial features, exhibit considerable variability. We hypothesise that analyzing this variability could provide qualitative insight into the severity due to the bacterial load for the specific subject under consideration. Specifically, we propose that displacements within the discriminative feature space correlate with underlying changes in pattern formation. This process can be conceptualized by comparing the dried patterns to the baseline control scenario (U-EC-00). A significant deviation in the resulting feature vector from this baseline suggests a substantial change in pattern formation 1 . Thus, the magnitude of displacement within this discriminative feature space is expected to correlate with varying degrees of bacterial concentration. This measured variation in the feature representations must reflect the alteration in pattern formation behavior induced by the presence of E. coli . Accordingly, cluster validity indices (CVIs) were evaluated on these discriminative feature vectors to compute this qualitative severity factor. Although an unsupervised approach would provide a more holistic model of morphological disruption, explicitly modeling the underlying pixel-level distribution is unfeasible given our relatively limited dataset and the significant diversity within subsets. Therefore, we utilize the features extracted from our classification model to establish a preliminary, discriminative baseline for these transitions, thereby facilitating a quantifiable severity metric. Although this approach may not capture the complete morphological picture, it serves as a correlative metric for assessing the impact of bacterial concentration through observed changes in pattern formation. Furthermore, evaluating diagnostic severity solely based on geometric features could provide a representation of relative categories (e.g., low, medium, or high) rather than an absolute interpretation based on their magnitudes; thereby, we term it a qualitative metric. Results and Discussion Insights from Pattern Evolution and SEM Microcharacterization Fig. 1 compares the temporal pattern evolution during evaporation with (insets B1–B3) and without (insets A1–A3) bacterial presence. Outer ring formation in an evaporating droplet with a pinned contact line is driven by radially outward capillary flow [ 78 ]. Dictated by this transport mechanism, bacterial mass is swept toward the contact line, resulting in a significantly thicker ring. Specifically, U-EC-11 exhibited increased peripheral deposition compared to U-EC-00 during pattern evolution, forming a denser and more intense edge band (inset B2) [ 79 ]. Furthermore, SEM micrographs of the final dried deposits, as presented in Fig. 2 , reveal that elevated bacterial concentrations correlate with an increased microcrystal count within the deposits, indicating enhanced nucleation activity. This enhancement can be attributed to extracellular polymeric substances (EPS) secreted by bacteria, which provide favorable binding sites for crystallizing ions [ 80 , 81 ]. However, it must be emphasized that these were broadly observed trends from our experiments. Urine composition being highly variable [ 77 , 82 ], these specific morphological interactions cannot be generalized across all scenarios. This challenge of generalizability in the observed patterns necessitated leveraging a machine learning framework for robust pattern recognition. Morphological Pattern Classification The ternary classification of morphological patterns formed by E. coli -laden dried sessile urine droplets, using the final stacking ensemble layer built on the attention-weighted fused vectors, achieved significant predictive performance. Fig. 5 presents the performance metrics along with the row-normalized confusion matrix. The model exhibited strong discriminative capability on the held-out test data, yielding an overall accuracy of approximately 86%, with macro-averaged precision, recall, and F1-scores in the range of 87 to 88%. A more detailed analysis through the class-wise metrics as depicted in Table 2 , supported by dimensionality reduction plots and the row-normalized confusion matrix, revealed that the predictive performance improved with an increase in the bacterial concentration. Specifically, the U-EC-11 class attained near-perfect predictions, whereas the U-EC-00 and U-EC-09 classes yielded comparatively lower performance. Download figure Open in new tab Figure 5: (A) The performance metrics and the (B) row-normalized confusion matrix for morphological pattern classification, based on the stacking ensemble predictions. View this table: View inline View popup Download powerpoint Table 2: Class-wise performance metrics based on the stacking ensemble predictions. Fig. 6 presents the Receiver Operating Characteristic (ROC) curves derived from the stacking ensemble predictions, alongside the probability distributions for the magnitudes of the attention-weighted feature fusion vectors. The ROC analysis demonstrated a progressive increase in the area under the curve (AUC) with increasing bacterial concentration, reflecting enhanced discriminative performance across the classes. Furthermore, the feature vector magnitude distributions revealed that U-EC-00 and U-EC-09 shared highly comparable profiles, whereas U-EC-11 maintained a distinct, well-separated distribution. This clear magnitude separation for U-EC-11 was consistent with its superior class-wise performance metrics, confirming its high separability relative to the other two classes. Conversely, the partially overlapping distributions observed for U-EC-00 and U-EC-09 provided complementary evidence aligning with their comparatively lower discriminative accuracy. Download figure Open in new tab Figure 6: (C) Receiver Operating Characteristic (ROC) curves based on the stacking ensemble classification and the (D) probability distributions for the magnitude of the attention-weighted fused feature vectors, as class-wise metrics. The two-dimensional PCA and the three-dimensional t-SNE embeddings of the L 2 normalized attention-weighted feature vectors presented in Fig. 7 visualize the class separability and the structural topology of the high-dimensional feature space. The PCA projection offers a low-dimensional representation that preserves the global variance of the features in the high-dimensional space. Within this projection, U-EC-11 exhibited strong overall separability, whereas U-EC-00 and U-EC-09 formed partially overlapping clusters, indicative of their comparatively lower discriminative ability. Conversely, the t-SNE embedding probabilistically preserves local neighborhood relationships, even though its resulting axes lack direct physical interpretation. This non-linear mapping effectively highlighted finer clustering patterns at the subset level. Specifi-cally, distinct sub-clusters emerged at each bacterial concentration level, revealing clear grouping across all three test subsets. Download figure Open in new tab Figure 7: (E) 2D Principal Component Analysis (PCA) plot, followed by (F) 3D t-SNE plot based on the L 2 normalized attention-weighted fused feature vectors. Qualitative Severity Estimation The qualitative severity estimation was conducted independently for each subset of the held-out test dataset by processing them through the trained pattern classification model to obtain the corresponding attention-weighted feature vectors. The severity factor was then quantified using cluster validity indices (CVIs) outlined in Table 3 . Computed from the extracted features, the indices evaluated the E. coli -laden classes (U-EC-09 and U-EC-11) against the baseline (U-EC-00) for a given subset. This subsection presents the results corresponding to a representative case (Subset 3 in Fig. 3 ). View this table: View inline View popup Download powerpoint Table 3: Comparison of CVI indices for the U-EC-09 and U-EC-11 feature-clusters relative to the baseline U-EC-00, evaluated based on the attention-weighted fused features. Generally, higher Silhouette scores, Calinski-Harabasz indices, and centroidal distances, coupled with lower Davies-Bouldin indices, signify highly distinctive clustering within the high-dimensional feature space. The severity factor was interpreted based on the hypothesis elaborated in the preceding section. While it is widely established that no single Cluster Validity Index (CVI) performs optimally across all scenarios [ 69 ], the Silhouette score ( SH ) was explicitly selected for its unique advantages over global, unbounded metrics like the Calinski-Harabasz and Davies-Bouldin indices. These advantages include its capacity for point-level assessment, its strictly bounded range [−1, 1], and its direct interpretability regarding both intra-cluster cohesion and inter-cluster separation. To compute the Severity Factor ( S ), we normalized this Silhouette Score ( SH ) using min-max scaling to yield a value within the range [0, 1]. Consequently, an S value closer to 0 indicates poor clustering, whereas a value approaching 1 denotes highly distinctive, robust class separation. The severity factor, computed as the normalized Silhouette score, was approximately 0.66 for U-EC-09 and 0.76 for U-EC-11 feature-clusters relative to the baseline (U-EC-00) for the chosen subset. Notably, the severity factor exhibited consistently moderate to high values across all analyzed subsets, showing a systematic, monotonic increase with bacterial concentration. This trend qualitatively indicated that higher E. coli concentrations yield a consistently elevated severity index across all tested subsets in this study. This observation aligns with the clinical bench-mark that an E. coli concentration of 10 5 CFU/mL or greater suggests a significant infection [ 83 ]. Therefore, these results provide compelling evidence of the qualitative impact of varying E. coli concentrations on dried pattern formation. Specifically, the consistently moderate to high severity factor, coupled with its monotonic increase relative to bacterial load, provides definitive preliminary evidence that it is a qualitative indicator of E. coli presence and concentration. Al-though comprehensive validation across diverse concentrations is necessary, this derived index is not meant to be considered absolute. Rather, it serves as a demonstration that image-based pattern analysis techniques can quantify biological variations, suggesting a valuable metric for diagnostic support. Conclusion In this study, we demonstrated a machine-learning-based approach for analyzing the morphological patterns of dried deposits formed by E. coli -laden sessile urine droplets, using a deep convolutional architecture-based attentive feature fusion approach, highlighting its potential to be scaled up into rapid, low-cost, and accessible clinical diagnostic applications. Our two-pronged objective involved: first, performing a morphological pattern classification using three deep convolutional backbones enhanced by an attentive feature fusion mechanism, followed by an ensemble classification layer. This was complemented by linear and non-linear dimensionality reduction maps for visualizing the extracted feature embeddings. Subsequently, we developed a novel qualitative severity estimation factor, defined as a function of geometric discriminative features, subsequently applying cluster validity indices (CVIs) to these attention-weighted fused feature vectors. This severity metric provided a foundational baseline for measuring the disruption in underlying morphological pattern formation due to bacterial presence. Ultimately, this quantifiable assessment can be potentially calibrated against clinical uropathogens, positioning our approach for potential extension into real-world, automated diagnostic systems. Critically, the inference time is computationally negligible once the global morphological features are extracted, meaning the to-tal analytical turnaround is limited solely by the physical processes of evaporation and optical imaging. To enhance the predictive generalization and robustness, our future efforts will involve experiments including a wider range of subjects and to lower bacterial concentrations, and can be further extended to include samples from real-world UTI-infected patients. This framework is po-tentially scalable for primary-care deployment. Integrating it into mobile or embedded platforms will transform its predictive capabilities into a practical, real-time diagnostic tool, delivering an accessible and cost-effective solution for resource-limited clinical environments. Data Availability All data produced in the present study are available upon reasonable request to the authors. CRediT authorship contribution statement Saptarshi Basu (S.B.): Conceptualization, Writing – review & editing, Resources, Funding acquisition. Dipshikha Chakravortty (D.C.): Conceptualization, Writing – review & editing. M Ashwin Ganesh (A.G.): Conceptualization, Project administration, Methodology, Formal analysis, Investigation, Software, Visualization, Validation, Supervision, Writing – original draft, Writing – review & editing. Sophia M (S.M.), Jason Joy Poopady (J.J.), Abdur Rasheed (A.R.), Durbar Roy (D.R.), Amey Nitin Agharkar (A.A), Kirti Parmar (K.P): Data curation, Methodology, Investigation. Visakh Vaikuntanathan (V.V.): Formal analysis. Declaration of competing interest The authors report no conflict of interest. Acknowledgments A.G. acknowledges SERC (Supercomputer Education and Research Center) for GPU support. S.B. acknowledges support from the INAE (Indian National Academy of Engineering) Chair Professorship and the Department of Science and Technology (DST) under the SUPRA scheme. We would like to express our sincere thanks to Mr. Anmol Singh, Mr. Mradul Namdeo, and Mr. Raghavendra K N from the Indian Institute of Science for their contribution to the experimental studies. Appendix A. Geometry Based Features: As Starting point Based on visual observations, the dried deposits formed by E. coli -laden sessile urine droplets, seemed to show variability in the underlying pattern formation at various concentrations of E. coli . As a first step to decode this pattern formation behavior, a geometry-based approach was performed as a starting point, with a pertinent question in mind: Could we identify and extract features that contain the signature of the specific class? As a preliminary analysis, a small set of the experimental microscopic images was considered, and certain geometry-based features were evaluated. A few of the estimated features studied include, Number of Particles: N , Areal number density of particles within the deposit footprint Area: N/A 0 , Total area of particles ( A tot ) normalized with the area of the deposit footprint area: A ∗ = A tot /A 0 , Total area of particles normalized with the maximum possible area within the image, A ∗∗ = A tot /A im , Total perimeter of particles( P tot ) normalized with the perimeter of the deposit footprint area: P ∗ = P tot /P 0 , Total perimeter of particles normalized with the maximum possible perimeter within the image, P ∗∗ = P tot /P im , Shape factor: . Based on these geometry-based features, most of the features showed a distinction between U-EC-00 and U-EC-11. The variation of the shape factor has been depicted in Fig. A.8. Though not completely evident, there seemed to be a potential and a tangible possibility for understanding the patterns with a comprehensive analysis. Motivated by this preliminary study and the supporting hypothesis, the evident way ahead was to step into unveiling the variation in pattern formation by employing deep learning-based techniques, utilizing their capacity to identify complex correlations to uncover patterns and relationships in the data that are not trivially perceptible to humans. Download figure Open in new tab Figure A.8 Variation of Shape factor Ψ = A tot /P 2 for certain chosen samples with presence of E.coli. Footnotes This version incorporates major revisions to enhance the robustness and clarity of the analysis, results, and interpretation. Significant changes include: title updated to reflect the revised scope; experimental image set updated to strengthen morphological pattern analysis; machine learning methodology revised to ensure rigorous implementation; physical insights from Scanning Electron Microscopy and temporal pattern evolution analysis added; results interpretation expanded to include Receiver Operating Characteristic curves, feature vector magnitude probability distributions, modified dimensionality reduction maps, and modified severity estimation table. ↵ 1 A preliminary geometry-based feature analysis performed to study the variation in pattern formation behavior has been provided in Appendix A, for interested readers. References [1]. ↵ G. Mancuso , A. Midiri , E. Gerace , M. Marra , S. Zummo , C. Biondo , Urinary tract infections: The current scenario and future prospects , Pathogens 12 ( 1 ) ( 2023 ) 142 . doi: 10.3390/pathogens12010142 . OpenUrl CrossRef PubMed [2]. ↵ M. Medina , E. Castillo-Pino , An introduction to the epidemiology and burden of uri-nary tract infections , Therapeutic Advances in Urology 11 ( 2019 ) 1756287219832172 . doi: 10.1177/1756287219832172 . OpenUrl CrossRef PubMed [3]. ↵ T. Bermudez , J. E. Schmitz , M. Boswell , R. M. Humphries , Novel technologies for the diagnosis of urinary tract infections , Clinical Biochemistry 118 ( 2023 ) 110684 . doi: 10.1016/j.clinbiochem.2023.110684 . OpenUrl CrossRef [4]. ↵ K. A. Fleming , S. Horton , M. L. Wilson , R. Atun , K. DeStigter , J. Flanigan , S. Sayed , P. Adam , B. Aguilar , S. Andronikou , C. Boehme , W. Cherniak , A. N. Cheung , B. Dahn , L. Donoso-Bach , T. Douglas , P. Garcia , S. Hussain , H. S. Iyer , M. Kohli , A. B. Labrique , L. M. Looi , J. G. Meara , J. Nkengasong , M. Pai , K. L. Pool , K. Ramaiya , L. Schroeder , D. Shah , R. Sullivan , B. S. Tan , K. Walia , The Lancet Commission on diagnostics: transforming access to diagnostics , The Lancet 398 ( 10315 ) ( 2021 ) 1997 – 2050 . doi: 10.1016/S0140-6736(21)00673-5 . OpenUrl CrossRef PubMed [5]. ↵ J. V. A. Franco , N. Meza , Advances in the diagnosis of urinary tract infection: A narrative review , Cureus (feb 2021 ). doi: 10.7759/cureus.13096 . OpenUrl CrossRef [6]. ↵ F. A. Mohammed , K. K. Tune , B. G. Assefa , M. Jett , S. Muhie , Medical image classifications using convolutional neural networks: A survey of current methods and statistical modeling of the literature , Diagnostics 13 ( 11 ) ( 2023 ) 1854 . doi: 10.3390/diagnostics13111854 . OpenUrl CrossRef PubMed [7]. ↵ J. Li , J. Chen , Y. Tang , C. Wang , B. A. Landman , S. Zhou , Transforming medical imaging with transformers? a comparative review of key properties, current progresses, and future perspectives , Medical Image Analysis 85 ( 2023 ) 102762 . doi: 10.1016/j.media.2023.102762 . OpenUrl CrossRef PubMed [8]. ↵ K. He , X. Zhang , S. Ren , J. Sun , Deep residual learning for image recognition, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2016 , pp. 770 – 778 . doi: 10.1109/CVPR.2016.90 . OpenUrl CrossRef [9]. ↵ K. Chaudhuri , R. Salakhutdinov M. Tan , Q. Le , Efficientnet: Rethinking model scaling for convolutional neural networks , in: K. Chaudhuri , R. Salakhutdinov (Eds.), Proceedings of the 36th International Conference on Machine Learning , Vol. 97 of Proceedings of Machine Learning Research, PMLR , 2019 , pp. 6105 – 6114 . URL https://proceedings.mlr.press/v97/tan19a.html [10]. ↵ C. Szegedy , W. Liu , Y. Jia , P. Sermanet , S. Reed , D. Anguelov , D. Erhan , V. Vanhoucke , A. Rabinovich , Going deeper with convolutions, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2015 , pp. 1 – 9 . doi: 10.1109/CVPR.2015.7298594 . OpenUrl CrossRef [11]. ↵ F. Zhuang , Z. Qi , K. Duan , D. Xi , Y. Zhu , H. Zhu , H. Xiong , Q. He , A comprehensive survey on transfer learning , Proceedings of the IEEE 109 ( 1 ) ( 2020 ) 43 – 76 . doi: 10.1109/JPROC.2020.3004555 . OpenUrl CrossRef [12]. ↵ A. Dosovitskiy , L. Beyer , A. Kolesnikov , D. Weissenborn , X. Zhai , T. Unterthiner , M. Dehghani , M. Minderer , G. Heigold , S. Gelly , J. Uszkoreit , N. Houlsby , An image is worth 16x16 words: Transformers for image recognition at scale , arXiv preprint arXiv:2010.11929 ( 2021 ). doi: 10.48550/arXiv.2010.11929 . OpenUrl CrossRef [13]. ↵ Z. Liu , Y. Lin , Y. Cao , H. Hu , Y. Wei , Z. Zhang , S. Lin , B. Guo , Swin transformer: Hierarchical vision transformer using shifted windows, in: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , 2021 , pp. 10012 – 10022 . doi: 10.1109/ICCV48922.2021.00986 . OpenUrl CrossRef [14]. ↵ Z. Tu , H. Talebi , H. Zhang , F. Yang , P. Milanfar , A. Bovik , Y. Li , Maxvit: Multi-axis vision transformer, in: European Conference on Computer Vision (ECCV) , 2022 , pp. 459 – 479 . doi:10.1007/978-3-031-20050-227. [15]. ↵ Z. Liu , H. Mao , C.-Y. Wu , C. Feichtenhofer , T. Darrell , S. Xie , A convnet for the 2020s, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2022 , pp. 11976 – 11986 . doi: 10.1109/CVPR52688.2022.01167 . OpenUrl CrossRef [16]. ↵ C. Szegedy , S. Ioffe , V. Vanhoucke , A. A. Alemi , Inception-v4, inception-resnet and the impact of residual connections on learning, in: Thirty-first AAAI conference on artificial intelligence , 2017 . doi: 10.1609/aaai.v31i1.11231 . OpenUrl CrossRef [17]. ↵ Z. Dai , H. Liu , Q. V. Le , M. Tan , Coatnet: Marrying convolution and attention for all data sizes , in: Advances in Neural Information Processing Systems (NeurIPS) , Vol. 34 , 2021 , pp. 3965 – 3977 . OpenUrl [18]. ↵ S. Mehta , M. Rastegari , Mobilevit: Light-weight, general-purpose, and mobile-friendly vision transformer, in: International Conference on Learning Representations (ICLR) , 2022 . URL https://openreview.net/forum?id=vh-i80-8E4 [19]. ↵ Z. Lu , H. Xie , C. Liu , Y. Zhang , Bridging the gap between vision transformers and convolutional neural networks on small datasets , in: Advances in Neural Information Processing Systems (NeurIPS) , Vol. 35 , 2022 , pp. 14663 – 14677 . OpenUrl [20]. ↵ X. Tang , The role of artificial intelligence in medical imaging research , BJR| Artificial Intelligence 1 ( 1 ) ( 2018 ) 20180025 . doi:10.1259/bjra.20180025. OpenUrl [21]. ↵ M. M. Zahoor , S. H. Khan , T. J. Alahmadi , T. Alsahfi , A. S. A. Mazroa , H. A. Sakr , S. Alqahtani , A. Albanyan , B. K. Alshemaimri , Brain tumor mri classification using a novel deep residual and regional cnn , Biomedicines 12 ( 7 ) ( 2024 ) 1395 . doi: 10.3390/biomedicines12071395 . OpenUrl CrossRef PubMed [22]. ↵ A. Esteva , B. Kuprel , R. A. Novoa , J. Ko , S. M. Swetter , H. M. Blau , S. Thrun , Dermatologist-level classification of skin cancer with deep neural networks , Nature 542 ( 7639 ) ( 2017 ) 115 – 118 . doi: 10.1038/nature21056 . OpenUrl CrossRef PubMed [23]. ↵ S. Azizi , L. Culp , J. Freyberg , B. Mustafa , S. Baur , S. Kornblith , T. Chen , et al. , Robust and data-efficient generalization of self-supervised machine learning for diagnostic imaging , Nature Biomedical Engineering 7 ( 2023 ) 756 – 779 . doi: 10.1038/s41551-023-01049-7 . OpenUrl CrossRef PubMed [24]. ↵ O. Ronneberger , P. Fischer , T. Brox , U-net: Convolutional networks for biomedical image seg-mentation , in: International Conference on Medical image computing and computer-assisted intervention , Springer , 2015 , pp. 234 – 241 . doi:10.1007/978-3-319-24574-428. [25]. ↵ J. Chen , Y. Lu , Q. Yu , X. Luo , E. Adeli , Y. Wang , L. Lu , A. L. Yuille , Y. Zhou , Transunet: Transformers make strong encoders for medical image segmentation , arXiv preprint arXiv:2102.04306 ( 2021 ). doi: 10.48550/arXiv.2102.04306 . OpenUrl CrossRef [26]. ↵ H. Cao , Y. Wang , J. Chen , D. Jiang , X. Zhang , Q. Tian , M. Wang , Swinunet: Unet-like pure transformer for medical image segmentation , in: European Conference on Computer Vision (ECCV) Workshops , Springer , 2022 . [27]. ↵ X. Wang , R. Girshick , A. Gupta , K. He , Attention mechanisms in computer vision: A survey , arXiv preprint arXiv:1709.04327 ( 2018 ). doi: 10.48550/arXiv.1709.04327 . OpenUrl CrossRef [28]. ↵ X. Li , W. Wang , X. Hu , J. Yang , Selective kernel networks, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2019 , pp. 510 – 519 . doi: 10.1109/CVPR.2019.00060 . OpenUrl CrossRef [29]. ↵ J. Hu , L. Shen , G. Sun , Squeeze-and-excitation networks, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2018 , pp. 7132 – 7141 . doi: 10.1109/CVPR.2018.00745 . OpenUrl CrossRef [30]. ↵ S. Woo , J. Park , J.-Y. Lee , I. S. Kweon , Cbam: Convolutional block attention module, in: Proceedings of the European Conference on Computer Vision (ECCV) , 2018 , pp. 3 – 19 . doi:10.1007/978-3-030-01234-21. [31]. ↵ Q. Wang , B. Wu , P. Zhu , P. Li , W. Zuo , Q. Hu , Eca-net: Efficient channel attention for deep convolutional neural networks , in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020 , pp. 11531 – 11539 . doi: 10.1109/CVPR42600.2020.01155 . OpenUrl CrossRef [32]. ↵ Y. Dai , F. Gieseke , S. Oehmcke , Y. Wu , K. Barnard , Attentional feature fusion, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , 2021 , pp. 3560 – 3569 . doi: 10.1109/WACV48630.2021.00360 . OpenUrl CrossRef [33]. ↵ S. Rajaraman , J. Siegelman , P. O. Alderson , L. R. Folio , S. K. Antani , G. R. Thoma , Feature fusion-based ensemble cnn learning for automated detection of pediatric pneumonia , Applied Sciences 10 ( 11 ) ( 2020 ) 3725 . doi: 10.3390/app10113725 . OpenUrl CrossRef [34]. ↵ R. J. Chen , M. Y. Lu , J. Wang , D. F. K. Williamson , S. J. Rodig , N. I. Lindeman , F. Mahmood , Pathomic fusion: An integrated framework for fusing histopathology and ge-nomic features for cancer diagnosis and prognosis , arXiv preprint arXiv:1811.08578 ( 2018 ). doi: 10.48550/arXiv.1811.08578 . OpenUrl CrossRef [35]. ↵ J. Qezelbash-Chamak , K. Hicklin , A hybrid learnable fusion of convnext and swin transformer for optimized image classification , IoT 6 ( 2 ) ( 2025 ) 30 . doi: 10.3390/iot6020030 . OpenUrl CrossRef [36]. ↵ M. I. Rahman , Fusion of vision transformer and convolutional neural network for explainable and efficient histopathological image classification in cyber-physical healthcare systems , Journal of Transformative Technologies and Sustainable Development 9 ( 2025 ) 8 . doi: 10.1007/s41314-025-00079-0 . OpenUrl CrossRef [37]. ↵ S. N. Shareef , V. Naralasetti , J. D. Bodapati , Adaptive fusion attention for enhanced classifica-tion and interpretability in medical imaging , Machine Vision and Applications 36 ( 3 ) ( 2025 ) 56 . doi: 10.1007/s00138-025-01665-0 . OpenUrl CrossRef [38]. ↵ M. Parsa , S. Harmand , K. Sefiane , Mechanisms of pattern formation from dried sessile drops , Advances in Colloid and Interface Science 254 ( 2018 ) 22 – 47 . doi: 10.1016/j.cis.2018.03.007 . OpenUrl CrossRef PubMed [39]. ↵ K. Sefiane , G. Duursma , A. Arif , Patterns from dried drops as a characterisation and healthcare diagnosis technique, potential and challenges: A review , Advances in Colloid and Interface Science 321 ( 2023 ) 103002 . doi: 10.1016/j.cis.2023.103002 . OpenUrl CrossRef PubMed [40]. ↵ A. Pal , A. Gope , A. Sengupta , Drying of bio-colloidal sessile droplets: Advances, applications, and perspectives , Advances in Colloid and Interface Science 314 ( 2023 ) 102870 . doi: 10.1016/j.cis.2023.102870 . OpenUrl CrossRef PubMed [41]. ↵ D. Brutin , B. Sobac , B. Loquet , J. Sampol , Pattern formation in drying drops of blood , Journal of Fluid Mechanics 667 ( 2011 ) 85 – 95 . doi:10.1017/jfm.2010.462. OpenUrl [42]. ↵ K. Sefiane , Patterns from drying drops , Advances in Colloid and Interface Science 206 ( 2014 ) 372 – 381 . doi: 10.1016/j.cis.2013.12.007 . OpenUrl CrossRef PubMed [43]. ↵ R. Chen , L. Zhang , D. Zang , W. Shen , Blood drop patterns: Formation and applications , Advances in Colloid and Interface Science 231 ( 2016 ) 1 – 14 . doi: 10.1016/j.cis.2016.03.004 . OpenUrl CrossRef PubMed [44]. ↵ D. Brutin , V. Starov , Recent advances in droplet wetting and evaporation , Chemical Society Reviews 47 ( 2 ) ( 2018 ) 558 – 585 . doi: 10.1039/c6cs00902f . OpenUrl CrossRef PubMed [45]. ↵ F. Giorgiutti-Dauphiné , L. Pauchard , Drying drops containing solutes: From hydrodynamical to mechanical instabilities , The European Physical Journal E 41 ( 3 ) ( 2018 ) 32 . doi:10.1140/epje/i2018-11641-5. OpenUrl [46]. ↵ F. R. Smith , D. Brutin , Wetting and spreading of human blood: Recent advances and applications , Current Opinion in Colloid & Interface Science 36 ( 2018 ) 78 – 83 . doi: 10.1016/j.cocis.2018.04.003 . OpenUrl CrossRef [47]. ↵ K. Sefiane , G. Duursma , M. A. Arif , Patterns from dried drops as a characterisation and healthcare diagnosis technique, potential and challenges: A review , Advances in Colloid and Interface Science 298 ( 2021 ) 102546 . doi: 10.1016/j.cis.2021.102546 . OpenUrl CrossRef PubMed [48]. ↵ R. Iqbal , A. Q. Shen , A. K. Sen , Understanding of the role of dilution on evaporative deposition patterns of blood droplets over hydrophilic and hydrophobic substrates , Journal of Colloid and Interface Science 579 ( 2020 ) 541 – 550 . doi: 10.1016/j.jcis.2020.04.109 . URL 10.1016/j.jcis.2020.04.109 OpenUrl CrossRef [49]. ↵ D. Roy , S. M. Joseph , K. K. Dewaangan , A. Rasheed , S. Jain , A. Singh , D. Chakravortty , S. Basu , Insights into the mechanics of pure and bacteria-laden sessile whole blood droplet evaporation , Journal of Fluid Mechanics 1021 ( 2025 ) A13 . doi: 10.1017/jfm.2025.10720 . OpenUrl CrossRef [50]. ↵ A. Pal , M. Yanagisawa , A. Gope , Time-lapse quantitative analysis of drying patterns and machine learning for classifying abnormalities in sessile blood droplets , medRxiv ( 2024 ). doi: 10.1101/2024.05.15.24307398 . OpenUrl Abstract / FREE Full Text [51]. ↵ O. Hegde , R. Chatterjee , A. Rasheed , D. Chakravortty , S. Basu , Multiscale vapor-mediated dendritic pattern formation and bacterial aggregation in complex respiratory biofluid droplets , Journal of Colloid and Interface Science 606 ( 2 ) ( 2022 ) 2011 – 2023 . doi: 10.1016/j.jcis.2021.09.158 . OpenUrl CrossRef [52]. ↵ R. Yang , N. Cao , Y. Yang , S. Guan , C. Yang , B. Situ , Y. Rui , H. Zhou , L. Zheng , Label-free urinary protein detection through machine learning analysis of single droplet evaporation patterns , Biosensors and Bioelectronics 211 ( 2022 ) 114389 . doi: 10.1016/j.bios.2022.114389 . OpenUrl CrossRef [53]. ↵ S. Sridharan , T. Bailey , A. Marcato , E. Simone , N. Watson , Artificial intelligence in colloid and interface science: Current research, challenges and future directions , Advances in Colloid and Interface Science 334 ( 2024 ) 103278 . doi: 10.1016/j.cis.2024.103278 . OpenUrl CrossRef [54]. ↵ O. Hegde , R. Chatterjee , D. Roy , V. Jaiswal , S. Jain , D. Chakravortty , S. Basu , Convolution neural networks for point-of-care diagnostics of bacterial infections in blood, in: A. De, A. K. Gupta, A. K. Runchal, D. Rakshit, A. Kushari (Eds.), Recent Developments in Power and Propulsion Applications, Green Energy and Technology, Springer , Singapore , 2025 , pp. 909 – 923 . doi:10.1007/978-981-96-8602-533. [55]. ↵ R. Demir , S. Koc , D. G. Ozturk , S. Bilir , H. İ. Özata , R. Williams , J. R. E. Christy , Y. Akkoc , İ. Tinay , C. Gunduz-Demir , D. Gozuacik , Artificial intelligence assisted patient blood and urine droplet pattern analysis for non-invasive and accurate diagnosis of bladder cancer , Scientific Reports 14 ( 2024 ). doi: 10.1038/s41598-024-52728-7 . OpenUrl CrossRef [56]. ↵ L. Shen , J. An , N. Wang , J. Wu , J. Yao , Y. Gao , Artificial intelligence and machine learning applications in urinary tract infections identification and prediction: a systematic review and meta-analysis , Frontiers in Public Health 11 ( 2023 ) 1120038 . doi:10.3389/fpubh.2023.1120038. OpenUrl [57]. ↵ J. M. Sheele , R. L. Campbell , D. D. Jones , Machine learning to predict bacteriuria in the emergency department , Diagnosis 10 ( 2 ) ( 2022 ) 101 – 107 . doi: 10.1515/dx-2022-0055 . OpenUrl CrossRef [58]. ↵ S. Farashi , H. E. Momtaz , Prediction of urinary tract infection using machine learning methods: a study for finding the most-informative variables , Scientific Reports 11 ( 1 ) ( 2021 ) 17315 . doi:10.1038/s41598-021-96791-2. OpenUrl PubMed [59]. ↵ P. M. C. Arches , A. R. Austero , A. J. A. Diaz , H. J. C. Taer , A. C. Fabregas , Detecting urinary tract infection (uti) thru analytes level using convolutional neural network and support vector machine , in: Proceedings of the 2021 3rd International Conference on Computer Science and Information Technology, ACM, 2021 , pp. 56 – 60 . [60]. ↵ J. Liou , K. Choi , G. Yang , J. Lee , T. Lee , K. Jeong , T. Im , W.-S. Lee , Y. Cho , A clinical microscopy dataset to develop a deep learning diagnostic test for urinary tract infection , Scientific Data 9 ( 1 ) ( 2022 ) 1 – 7 . doi:10.1038/s41597-022-01254-6. OpenUrl PubMed [61]. ↵ R. Iriya , B. Braswell , M. Mo , F. Zhang , S. E. Haydel , S. Wang , Deep learning-based culture-free bacteria detection in urine using large-volume microscopy , Biosensors 14 ( 4 ) ( 2024 ) 280 . doi:/10.3390/bios14040280. OpenUrl CrossRef PubMed [62]. ↵ D. Clayson , D. Wild , H. Doll , K. Keating , K. Gondek , Validation of a patient-administered questionnaire to measure the severity and bothersomeness of lower urinary tract symptoms in uncomplicated urinary tract infection (uti): The uti symptom assessment questionnaire , Health and Quality of Life Outcomes 4 ( 2006 ) 6 . doi: 10.1186/1477-7525-4-6 . OpenUrl CrossRef PubMed [63]. ↵ A. D. Hay , K. Birnie , J. R. Benger , J. A. Sterne , B. Barr , B. Delaney , N. Eagles , M. Fletcher , G. Funston , D. Gaunt , W. Hollingworth , K. Hood , R. Howe , J. Macleod , M. Moore , P. Muir , S. Noble , S. Piper , C. Powell , K. Rumsby , F. Thomas , M. Thompson , P. Whiting , V. Wilson , M. Wootton , C. C. Butler , Development and validation of a clinical rule for the diagnosis of urinary tract infection in children under five in primary care: a prospective cohort study , The British Journal of General Practice 66 ( 648 ) ( 2016 ) e486 – e497 . doi:10.3399/bjgp16X685881. OpenUrl [64]. ↵ R. De Rosa , S. Grosso , G. Bruschetta , F. Stano , I. Lorenzon , A. Fenza , C. Gallo , M. D’Avino , A. Orsini , P. Sglavo , Development of a uti-risk score based on the integration of flow cytometry and urinalysis to improve the accuracy of uti diagnosis , Journal of Clinical Medicine 9 ( 5 ) ( 2020 ) 1563 . doi: 10.3390/jcm9051563 . OpenUrl CrossRef [65]. ↵ R. Dubes , A. K. Jain , Validity studies in clustering methodologies , Pattern Recognition 11 ( 4 ) ( 1979 ) 235 – 254 . doi:10.1016/0031-3203(79)90012-9. OpenUrl CrossRef Web of Science [66]. ↵ P. J. Rousseeuw , Silhouettes: A graphical aid to the interpretation and validation of cluster analysis , Journal of Computational and Applied Mathematics 20 ( 1987 ) 53 – 65 . doi: 10.1016/0377-0427(87)90125-7 . OpenUrl CrossRef PubMed Web of Science [67]. ↵ D. L. Davies , D. W. Bouldin , A cluster separation measure , IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-1 ( 2 ) ( 1979 ) 224 – 227 . doi: 10.1109/TPAMI.1979.4766909 . OpenUrl CrossRef PubMed Web of Science [68]. ↵ T. Caliński , J. Harabasz , A dendrite method for cluster analysis , Communications in Statistics 3 ( 1 ) ( 1974 ) 1 – 27 . doi: 10.1080/03610927408827101 . OpenUrl CrossRef [69]. ↵ O. Arbelaitz , I. Gurrutxaga , J. Muguerza , J. M. Pérez , I. Perona , An extensive com-parative study of cluster validity indices , Pattern Recognition 46 ( 1 ) ( 2013 ) 243 – 256 . doi: 10.1016/j.patcog.2012.07.021 . OpenUrl CrossRef [70]. ↵ C. Cortes , V. Vapnik , Support-vector networks , Machine Learning 20 ( 3 ) ( 1995 ) 273 – 297 . doi: 10.1007/BF00994018 . OpenUrl CrossRef Web of Science [71]. ↵ L. Breiman , Random forests , Machine Learning 45 ( 1 ) ( 2001 ) 5 – 32 . doi: 10.1023/A:1010933404324 . OpenUrl CrossRef PubMed [72]. ↵ R. O. Duda , P. E. Hart , Pattern classification and scene analysis , Wiley , 1973 . [73]. ↵ T. M. Cover , P. E. Hart , Nearest Neighbor Pattern Classification , Vol. 13, IEEE Transactions on Information Theory, 1967, the k-nearest neighbor (KNN) classifier assigns a class label to a sample based on the majority class among its k closest samples in the feature space. [74]. ↵ D. R. Cox , The regression analysis of binary sequences , Journal of the Royal Statistical Society: Series B (Methodological ) 20 ( 2 ) ( 1958 ) 215 – 232 . doi: 10.1111/j.2517-6161.1958.tb00292.x . OpenUrl CrossRef [75]. ↵ I. T. Jolliffe , Principal component analysis, Springer Series in StatisticsPrincipal Component Analysis (PCA) is a linear dimensionality reduction technique that projects high-dimensional data onto a low-dimensional (e.g., 2D) subspace while preserving maximum variance . ( 1986 ). [76]. ↵ L. van der Maaten , G. Hinton , Visualizing data using t-sne , Journal of Machine Learning Research 9 ( 2008 ) 2579 – 2605 . doi:10.5555/1390681.1390693. OpenUrl [77]. ↵ S. Bouatra , F. Aziat , R. Mandal , A. C. Guo , M. R. Wilson , C. Knox , T. C. Bjorndahl , R. Krishnamurthy , F. Saleem , P. Liu , et al. , The human urine metabolome , PloS one 8 ( 9 ) ( 2013 ) e73076 . doi: 10.1371/journal.pone.0073076 . OpenUrl CrossRef PubMed [78]. ↵ R. D. Deegan , O. Bakajin , T. F. Dupont , G. Huber , S. R. Nagel , T. A. Witten , Capillary flow as the cause of ring stains from dried liquid drops , Nature 389 ( 6653 ) ( 1997 ) 827 – 829 . doi: 10.1038/39827 . OpenUrl CrossRef [79]. ↵ S. Majee , A. R. Chowdhury , R. Pinto , A. Chattopadhyay , A. N. Agharkar , D. Chakravortty , S. Basu , Spatiotemporal evaporating droplet dynamics on fomites enhances long term bacterial pathogenesis , Communications Biology 4 ( 1 ) ( 2021 ) 1173 . doi: 10.1038/s42003-021-02711-z . OpenUrl CrossRef PubMed [80]. ↵ D. N. Azulay , R. Abbasi , I. Ben Simhon Ktorza , S. Remennik , M. Amarendar Reddy , L. Chai , Biopolymers from a bacterial extracellular matrix affect the morphology and structure of calcium carbonate crystals , Crystal Growth & Design 19 ( 7 ) ( 2019 ) 4180 – 4187 . doi: 10.1021/acs.cgd.8b00888 . OpenUrl CrossRef [81]. ↵ C. Ercole , P. Bozzelli , F. Altieri , P. Cacchio , M. Del Gallo , Calcium carbonate mineralization: Involvement of extracellular polymeric materials isolated from calcifying bacteria , Microscopy and Microanalysis 13 ( 1 ) ( 2007 ) 42 – 50 . doi:10.1017/S1431927607070125. OpenUrl [82]. ↵ E. J. Saude , B. D. Sykes , Standardizing the experimental conditions for using urine in nmr-based metabolomic studies with a particular focus on diagnostic studies: a review , Metabolomics 3 ( 1 ) ( 2007 ) 19 – 27 . doi:10.1007/s11306-006-0049-1. OpenUrl CrossRef Web of Science [83]. ↵ E. H. Kass , Asymptomatic infections of the urinary tract , Transactions of the Association of American Physicians 69 ( 1956 ) 56 – 64 . OpenUrl PubMed Web of Science View the discussion thread. Back to top Previous Next Posted May 11, 2026. Download PDF Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Machine Learning-based Morphological Pattern Analysis of Dried Deposits formed by E. coli-laden Sessile Urine Droplets for Point-of-Care Diagnostics Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Machine Learning-based Morphological Pattern Analysis of Dried Deposits formed by E. coli -laden Sessile Urine Droplets for Point-of-Care Diagnostics M Ashwin Ganesh , Sophia M , Jason Joy Poopady , Abdur Rasheed , Durbar Roy , Amey Nitin Agharkar , Visakh Vaikuntanathan , Kirti Parmar , Dipshikha Chakravortty , Saptarshi Basu medRxiv 2025.10.14.25337322; doi: https://doi.org/10.1101/2025.10.14.25337322 Share This Article: Copy Citation Tools Machine Learning-based Morphological Pattern Analysis of Dried Deposits formed by E. coli -laden Sessile Urine Droplets for Point-of-Care Diagnostics M Ashwin Ganesh , Sophia M , Jason Joy Poopady , Abdur Rasheed , Durbar Roy , Amey Nitin Agharkar , Visakh Vaikuntanathan , Kirti Parmar , Dipshikha Chakravortty , Saptarshi Basu medRxiv 2025.10.14.25337322; doi: https://doi.org/10.1101/2025.10.14.25337322 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Infectious Diseases (except HIV/AIDS) Subject Areas All Articles Addiction Medicine (569) Allergy and Immunology (863) Anesthesia (300) Cardiovascular Medicine (4442) Dentistry and Oral Medicine (444) Dermatology (383) Emergency Medicine (609) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1510) Epidemiology (15230) Forensic Medicine (30) Gastroenterology (1126) Genetic and Genomic Medicine (6608) Geriatric Medicine (668) Health Economics (998) Health Informatics (4542) Health Policy (1370) Health Systems and Quality Improvement (1613) Hematology (543) HIV/AIDS (1266) Infectious Diseases (except HIV/AIDS) (15923) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (147) Nephrology (668) Neurology (6607) Nursing (346) Nutrition (999) Obstetrics and Gynecology (1146) Occupational and Environmental Health (957) Oncology (3336) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (664) Pediatrics (1693) Pharmacology and Therapeutics (692) Primary Care Research (712) Psychiatry and Clinical Psychology (5448) Public and Global Health (9237) Radiology and Imaging (2201) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1196) Rheumatology (596) Sexual and Reproductive Health (714) Sports Medicine (530) Surgery (712) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a018a2618eb41b23',t:'MTc3OTc1NTAzOA=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();