SYSTEM AND METHOD FOR WHITE BLOOD CELL CLASSIFICATION UTILIZING PRE-TRAINED DEEP NEURAL NETWORKS

Abstract

ABSTRACT “SYSTEM AND METHOD FOR WHITE BLOOD CELL CLASSIFICATION UTILIZING PRE-TRAINED DEEP NEURAL NETWORKS” The present invention provides a system and a method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques. In this study, a computer-aided diagnosis system was proposed that utilizes pre-trained networks for accurate classification of white blood cells. The dataset used in this study is sourced from Kaggle, and the classification process is performed without image segregation or feature extraction techniques. Pre-trained series networks are employed for classification, and a classification accuracy of 99% is achieved using the Inception-ResNet-v2 network with the Adam optimizer. The comparative analysis aids in understanding the performance and suitability of different network architectures for white blood cell classification tasks. Figure 1

Specification

Description:TECHNICAL FIELD
[0001] The present invention relates to the field of artificial intelligence and automated systems, and more particularly, the present invention relates to the system and method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques.
BACKGROUND ART
[0002] The following discussion of the background of the invention is intended to facilitate an understanding of the present invention. However, it should be appreciated that the discussion is not an acknowledgment or admission that any of the material referred to was published, known, or part of the common general knowledge in any jurisdiction as of the application's priority date. The details provided herein the background if belongs to any publication is taken only as a reference for describing the problems, in general terminologies or principles or both of science and technology in the associated prior art.
[0003] The human body is a complex system of cells and organs that work together to maintain balance. Blood, a crucial component, consists of plasma, white blood cells (WBCs), red blood cells (RBCs), and platelets. WBCs are important for immune defense, and accurate identification and classification of these cells are essential for diagnosing and treating various medical conditions.
[0004] There are five main types of WBCs: neutrophils, eosinophils, basophils, lymphocytes, and monocytes. Each type serves a unique function, and changes in their concentrations can indicate different diseases or disorders. For example, an increase in lymphocytes may indicate a viral illness, while an increase in neutrophils may suggest a bacterial infection.
[0005] Traditionally, hematologists manually classify WBCs using a microscope, which is a time-consuming process. However, automated systems using computer vision and machine learning have been developed to speed up the process, reduce errors, and provide objective results.
[0006] Deep neural networks (DNNs) are a type of artificial neural network with multiple layers that can learn complex patterns from large datasets. Transfer learning, a popular method, involves using pre-trained DNNs on large datasets and fine-tuning them for a new task with a smaller dataset. This approach is particularly useful in medical diagnosis, where labeled data may be limited.
[0007] In this paper, the objective is to compare the performance of different transfer learning approaches for WBC classification. Six DNN architectures (VGG-16, MobileNet-v2, Xception, DenseNet-201, Inceptionv3, and Inception-ResNet-v2) will be evaluated based on accuracy, precision, f1-score, support, and recall. Additionally, the impact of specific hyperparameters, such as learning rate, batch size, and number of epochs, will be investigated.
[0008] A publicly available dataset of 12,500 blood cell images representing five different classes of WBCs will be used. The dataset will be divided into training, validation, and testing sets for model training, hyperparameter adjustment, and preventing overfitting. The models' performance will be evaluated using the testing set.
[0009] In light of the foregoing, there is a need for System and method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques that overcomes problems prevalent in the prior art associated with the traditionally available method or system, of the above-mentioned inventions that can be used with the presented disclosed technique with or without modification.
[0010] All publications herein are incorporated by reference to the same extent as if each individual publication or patent application were specifically and individually indicated to be incorporated by reference. Where a definition or use of a term in an incorporated reference is inconsistent or contrary to the definition of that term provided herein, the definition of that term provided herein applies, and the definition of that term in the reference does not apply.
OBJECTS OF THE INVENTION
[0011] The principal object of the present invention is to overcome the disadvantages of the prior art by providing system and method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques.
[0012] Another object of the present invention is to provide system and method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques that utilizes deep learning networks for automated WBC classification without the need for time-consuming methods like segmentation and feature extraction.
[0013] Another object of the present invention is to provide system and method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques that evaluates the performance of six DNN architectures for WBC classification.
[0014] Another object of the present invention is to provide system and method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques that compares the classification performance of six pre-trained DNNs on the BCCD dataset using two optimization algorithms: SGDM and ADAM.
[0015] The foregoing and other objects of the present invention will become readily apparent upon further review of the following detailed description of the embodiments as illustrated in the accompanying drawings.
SUMMARY OF THE INVENTION
[0016] The present invention relates to system and method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques.
[0017] In this paper, white blood cells (WBC) are classified using pre-trained deep CNNs without the need for image segmentation or manually extracted image attributes. Figure 1 (ref. Figure 1) provides a schematic depiction of the proposed categorization system. The blood smear images in the dataset are first resized to fit the pre-trained CNN's image input size, and then these images are processed by the CNN, allowing the network to automatically collect the features needed for classification during training. The categorization of the images into lymphocyte, monocyte, neutrophil, and eosinophil is subsequently carried out by the CNN's output layer, utilizing the learned characteristics. The Kaggle dataset was used for all of the categorization trials in the study.
[0018] The experiment utilized GPU and TPU accelerators for training the model on Kaggle. Accelerators help speed up training and handle large datasets. The initial learning rate was set to 0.000001 for fine adjustments to the model's weights. Two optimizers, Adam and SGD, were used to update the model's weights based on gradients. The hardware configuration included an Intel Core i5 processor with 4 cores and 16 GB of RAM. The training process lasted for 20 epochs, with a batch size of 32.
[0019] The dataset used was obtained from Kaggle Blood Cell Images, containing 12,444 augmented images of blood cells divided into four types. Preprocessing steps included data cleaning, resizing images to 128x128 pixels, and applying a median filter for noise reduction. Augmentation techniques like rotations flips, and shears were used to increase the dataset size. The dataset was split into 80% for training and 20% for testing (10% validation, 10% testing).
[0020] For object detection, additional preprocessing steps were performed. This included adding padding to improve cell detection at the image edges, thresholding to isolate target cells, erosion and dilation for noise removal and smoothing, contour detection, extracting bounding box coordinates, and masking the image to extract only the cell. The images were resized to a fixed size.
[0021] While the invention has been described and shown with reference to the preferred embodiment, it will be apparent that variations might be possible that would fall within the scope of the present invention.
BRIEF DESCRIPTION OF DRAWINGS
[0022] So that the manner in which the above-recited features of the present invention can be understood in detail, a more particular description of the invention, briefly summarized above, may have been referred by embodiments, some of which are illustrated in the appended drawings. It is to be noted, however, that the appended drawings illustrate only typical embodiments of this invention and are therefore not to be considered limiting of its scope, for the invention may admit to other equally effective embodiments.
[0023] These and other features, benefits, and advantages of the present invention will become apparent by reference to the following text figure, with like reference numbers referring to like structures across the views, wherein:
[0024] Figure 1: Proposed Fine-Tuned Model, in accordance with an exemplary embodiment of the present invention.
[0025] Figure 2: Classification Steps.
DETAILED DESCRIPTION OF THE INVENTION
[0026] While the present invention is described herein by way of example using embodiments and illustrative drawings, those skilled in the art will recognize that the invention is not limited to the embodiments of drawing or drawings described and are not intended to represent the scale of the various components. Further, some components that may form a part of the invention may not be illustrated in certain figures, for ease of illustration, and such omissions do not limit the embodiments outlined in any way. It should be understood that the drawings and the detailed description thereto are not intended to limit the invention to the particular form disclosed, but on the contrary, the invention is to cover all modifications, equivalents, and alternatives falling within the scope of the present invention as defined by the appended claim.
[0027] As used throughout this description, the word "may" is used in a permissive sense (i.e. meaning having the potential to), rather than the mandatory sense, (i.e. meaning must). Further, the words "a" or "an" mean "at least one" and the word "plurality" means "one or more" unless otherwise mentioned. Furthermore, the terminology and phraseology used herein are solely used for descriptive purposes and should not be construed as limiting in scope. Language such as "including," "comprising," "having," "containing," or "involving," and variations thereof, is intended to be broad and encompass the subject matter listed thereafter, equivalents, and additional subject matter not recited, and is not intended to exclude other additives, components, integers, or steps. Likewise, the term "comprising" is considered synonymous with the terms "including" or "containing" for applicable legal purposes. Any discussion of documents, acts, materials, devices, articles, and the like are included in the specification solely for the purpose of providing a context for the present invention. It is not suggested or represented that any or all these matters form part of the prior art base or were common general knowledge in the field relevant to the present invention.
[0028] In this disclosure, whenever a composition or an element or a group of elements is preceded with the transitional phrase "comprising", it is understood that we also contemplate the same composition, element, or group of elements with transitional phrases "consisting of", "consisting", "selected from the group of consisting of, "including", or "is" preceding the recitation of the composition, element or group of elements and vice versa.
[0029] The present invention is described hereinafter by various embodiments with reference to the accompanying drawing, wherein reference numerals used in the accompanying drawing correspond to the like elements throughout the description. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiment set forth herein. Rather, the embodiment is provided so that this disclosure will be thorough and complete and will fully convey the scope of the invention to those skilled in the art. In the following detailed description, numeric values and ranges are provided for various aspects of the implementations described. These values and ranges are to be treated as examples only and are not intended to limit the scope of the claims. In addition, several materials are identified as suitable for various facets of the implementations. These materials are to be treated as exemplary and are not intended to limit the scope of the invention.
[0030] The present invention relates to system and method for white blood cell classification utilizing pre-trained deep neural networks and transfer learning techniques.
[0031] The experiment utilized GPU and TPU accelerators for training the model on Kaggle. Accelerators help speed up training and handle large datasets. The initial learning rate was set to 0.000001 for fine adjustments to the model's weights. Two optimizers, Adam and SGD, were used to update the model's weights based on gradients. The hardware configuration included an Intel Core i5 processor with 4 cores and 16 GB of RAM. The training process lasted for 20 epochs, with a batch size of 32.
[0032] The dataset used was obtained from Kaggle Blood Cell Images, containing 12,444 augmented images of blood cells divided into four types. Preprocessing steps included data cleaning, resizing images to 128x128 pixels, and applying a median filter for noise reduction. Augmentation techniques like rotations flips, and shears were used to increase the dataset size. The dataset was split into 80% for training and 20% for testing(10% validation, 10% testing).
[0033] For object detection, additional preprocessing steps were performed. This included adding padding to improve cell detection at the image edges, thresholding to isolate target cells, erosion and dilation for noise removal and smoothing, contour detection, extracting bounding box coordinates, and masking the image to extract only the cell. The images were resized to a fixed size.
[0034] Pre-trained Models: Transfer learning is used in this experiment with six high-performing models: VGG-16, MobileNet-v2, Xception, DenseNet-201, Inceptionv3, and Inception-ResNet-v2. These models are pre-trained on a large dataset and known for their exceptional performance in computer vision tasks. By leveraging their pre-trained weights, training time is reduced, and model accuracy is improved. The models have been fine-tuned on the specific dataset used in this experiment to enhance their performance further.
[0035] InceptionResNetV2 is a deep CNN architecture that combines the Inception module and residual connections from ResNet. It achieves state-of-the-art performance on computer vision tasks by capturing intricate image features effectively through over 50 layers.
[0036] DenseNet201 is CNN architecture with dense connectivity between layers. It encourages feature reuse and information flow by connecting each layer to every subsequent layer in a feed-forward manner. With over 200 layers, it captures local and global image features effectively.
[0037] Inceptionv3 is a popular CNN architecture developed by Google, aiming for improved accuracy and efficiency. It incorporates the Inception module with parallel convolutional filters of different sizes and utilizes optimizations like factorizing convolutions and batch normalization.
[0038] Xception is an advanced CNN architecture inspired by Inception modules. It replaces standard convolutions with depthwise separable convolutions, capturing spatial and cross-channel correlations efficiently. It achieves state-of-the-art results on image classification and suits resource-constrained environments.
[0039] MobileNet-v2 is a lightweight CNN architecture designed for mobile and embedded devices. It focuses on efficiency with depthwise separable convolutions, inverted residual blocks, and linear bottlenecks. It maintains high accuracy while being suitable for real-time processing and low memory footprint.
[0040] VGG-16 is an influential CNN architecture developed by the VGG group. It consists of 16 layers, employing 3x3 convolutional filters throughout the network. VGG-16 achieved remarkable performance on the ImageNet challenge and has served as a benchmark for subsequent CNN architectures.
[0041] Optimization Methods: Adam is an adaptive optimization algorithm that combines Momentum and RMSProp techniques. It adjusts the learning rate for each parameter based on the first and second moments of the gradients, allowing it to effectively handle different gradient types, including sparse gradients. It is popular for optimizing deep neural networks due to its quick convergence and versatility.
[0042] SGD is a fundamental optimization algorithm that updates model parameters using the loss function's gradient. It performs parameter updates individually for each training example, making it efficient for large datasets. However, SGD may have slower convergence and oscillate in certain scenarios.
[0043] Performance Evaluation Matrix: Classification models are assessed using metrics like accuracy, precision, recall, F1-score, and support. These measures provide valuable insights into the model's performance in categorizing classes. Let's understand each metric and its formula:
- True Positive: Correctly classified as positive instances.
- True Negative: Correctly classified negative instances.
- False Positive: Incorrectly classified negative instances as positive.
- False Negative: Incorrectly classified positive instances as negative.
[0044] Accuracy: Accuracy measures the overall correctness of the model's predictions by comparing the total correct predictions to the total predictions made.
[0045] Accuracy = (True Positives + True Negatives) / (True Positives + True Negatives + False Positives + False Negatives)
[0046] Precision: Precision evaluates the proportion of correctly predicted positive cases out of all instances predicted as positive. It measures the model's ability to avoid false positive errors.
[0047] Precision = True Positives / (True Positives + False Positives)
[0048] Recall: Recall (Sensitivity/True Positive Rate) calculates the percentage of correctly predicted positive cases out of all actual positive instances. It assesses the model's accuracy in identifying positive cases.
[0049] Recall = True Positives / (False Negatives + True Positives)
[0050] F1-score: The F1-score is the harmonic mean of precision and recall. It provides a balanced measure of accuracy and recall, considering both false positives and false negatives. It is useful for imbalanced datasets.
[0051] F1-score = 2 * (Precision * Recall) / (Precision + Recall)
[0052] Support: Support indicates the frequency of each class in the dataset. It helps understand the significance of performance metrics and provides information about class distribution. Load and Preprocess Data: The dataset contains images of different blood cell types. Images are loaded, converted to RGB format, and preprocessed using techniques like padding, thresholding, and morphological operations to enhance and extract blood cells. Data Splitting: The dataset is split into training, validation, and testing sets to evaluate the model's performance. The split is done in a layered manner to maintain the class distribution in each set. Model Architecture: Pre-trained CNN models like VGG-16, MobileNet-v2, Xception, DenseNet-201, Inceptionv3, and Inception-ResNet-v2 are used. Specific layers are frozen, and additional layers are added for customization. Dropout regularization is applied to prevent overfitting. Compile and Train Model: The model is compiled with a loss function, optimizer, and evaluation metric. Training is done in mini-batches, and the model's performance is evaluated on the validation set. Call backs are used to save the best weights, stop training early, and adjust the learning rate. Model Evaluation and Analysis: The trained model is evaluated on the testing, validation, and training sets using metrics like accuracy and loss. Classification metrics and a confusion matrix provide a detailed evaluation of the model's performance. Visualization and Reporting: Plots show accuracy/loss trends, class distribution, and example images. A confusion matrix visualizes predictions and errors. Model Saving and Final Analysis: The trained model is saved for future use. Its performance is analyzed on the testing set, and a confusion matrix identifies misclassifications and patterns.
[0053] The proposed study focuses on using transfer learning and pre-trained models for automatically identifying white blood cells (WBCs) from the BCCD dataset. Unlike computationally demanding methods like image segmentation or manual feature extraction, this study solely relies on pre-trained CNNs for classification. The research compares six capable pre-trained models and two optimization approaches. Among these models, the Inception ResNetV2 stands out as the best performer, achieving the highest accuracy on the WBCs dataset. The model is trained using the Adam optimizer, and it achieves an impressive classification accuracy of 98.51% in just 20 epochs for the four WBC classes present in the dataset. This high accuracy demonstrates the effectiveness of deep learning models for WBC classification and indicates the potential of transfer learning in medical image analysis tasks. Considering both time complexity and classification accuracy, the Inception ResNetV2 model with the Adam optimizer proves to be the most favorable choice for the automated identification of white blood cells. Its use of transfer learning from a pre-trained model allows for efficient training and reduces the need for computationally demanding tasks like manual feature extraction. However, it's important to note that the study acknowledges the possibility of further data collection to enhance training, which could potentially improve performance even further.
[0054] In our future research on automated white blood cell (WBC) classification, we have several avenues to explore. Firstly, we aim to categorize additional WBC subtypes to enhance the comprehensiveness of the classification system. Secondly, we will develop techniques to handle challenging WBC images with multi-celled, overlapping, or obscured cells, thereby improving the models' generalizability for real-world scenarios. Thirdly, we will investigate different deep learning architectures beyond those previously discussed, with the aim of finding more effective models for WBC classification. Additionally, we plan to expand the range of WBC datasets used for evaluation and validation, incorporating diverse datasets to gain a comprehensive understanding of the proposed methods' performance and generalizability.
[0055] Various modifications to these embodiments are apparent to those skilled in the art from the description and the accompanying drawings. The principles associated with the various embodiments described herein may be applied to other embodiments. Therefore, the description is not intended to be limited to the 5 embodiments shown along with the accompanying drawings but is to be providing the broadest scope consistent with the principles and the novel and inventive features disclosed or suggested herein. Accordingly, the invention is anticipated to hold on to all other such alternatives, modifications, and variations that fall within the scope of the present invention and appended claims. , Claims:CLAIMS
We Claim:
1) A system for classifying white blood cells using transfer learning and pre-trained deep neural networks, the system comprising:
- a data acquisition module for obtaining an image dataset of blood cells;
- a preprocessing module configured to clean, resize, and filter the images for noise reduction, wherein each image is preprocessed to a specified resolution and augmented through rotation, flipping, and shearing techniques to enhance the dataset size;
- an object detection module that isolates individual white blood cells within each image by:
- applying padding to account for cells near image edges,
- applying thresholding techniques to isolate target cells,
- performing morphological operations including erosion and dilation for noise reduction,
- detecting contours to identify cells,
- extracting bounding box coordinates, and
- masking each image to retain only the white blood cells;
- a deep neural network module incorporating multiple pre-trained convolutional neural network (CNN) architectures selected from a group consisting of VGG-16, MobileNet-v2, Xception, DenseNet-201, Inceptionv3, and Inception-ResNet-v2, each fine-tuned using transfer learning with training data from the preprocessed dataset;
- an optimization module configured to train the neural network model using optimizers, selected from Adam and Stochastic Gradient Descent (SGD), to adjust weights for improved classification accuracy;
- a classification module configured to classify white blood cells into distinct classes based on features learned by the neural network model;
- an evaluation module configured to assess the performance of the neural network model using metrics including accuracy, precision, recall, and F1-score.
2) A method for automated classification of white blood cells using transfer learning and pre-trained deep neural networks, the method comprising the steps of:
- acquiring a dataset of blood cell images and preprocessing the images by cleaning, resizing to a specified resolution, and augmenting to create an enhanced dataset;
- isolating individual white blood cells within each image by:
- applying padding, thresholding, and morphological operations to improve detection,
- detecting contours to locate cells,
- extracting bounding box coordinates, and
- masking the images to retain only the white blood cells;
- selecting one or more pre-trained CNN architectures from VGG-16, MobileNet-v2, Xception, DenseNet-201, Inceptionv3, and Inception-ResNet-v2;
- performing transfer learning on the selected CNN architectures by fine-tuning them with training data from the preprocessed dataset;
- training the CNN model using an optimizer selected from Adam or SGD to update weights based on gradients and achieve optimal classification performance;
- evaluating the performance of the trained CNN model using metrics including accuracy, precision, recall, F1-score, and support;
- classifying the white blood cells within the images based on features learned by the trained CNN model.

Documents