MRI-Enhanced Metastatic Ovarian Tumor Detection: Leveraging Enhanced 3D CNN and Data Augmentation for Exceptional Accuracy

doi:10.21203/rs.3.rs-4854264/v1

Download PDF

Research Article

MRI-Enhanced Metastatic Ovarian Tumor Detection: Leveraging Enhanced 3D CNN and Data Augmentation for Exceptional Accuracy

https://doi.org/10.21203/rs.3.rs-4854264/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background/Aims: Metastatic Ovarian Tumor is a severe condition that can significantly impact the life span and quality of life of affected individuals. Common symptoms include hormonal imbalances, digestive system issues, pelvic pain, fertility problems, and depression. Accurate and early detection is essential for improving patient outcomes. This research aims to develop a more effective diagnostic tool using MRI and 3D Convolutional Neural Networks (CNN) to enhance early detection and diagnosis of metastatic ovarian tumors.

Materials and Methods: This study leverages the power of 3D Convolutional Neural Networks (CNN) to analyze MRI scans for the detection of metastatic ovarian tumors. The proposed model employs a 3D CNN architecture, known for its effectiveness in image classification tasks. Existing approaches using 2D CNNs often fail to capture the spatial and temporal features of MRI scans, leading to information loss. To improve model performance, data augmentation techniques such as random cropping, resizing, and spatial deformation were integrated. The model was tested with the Ovarian Bevacizumab Response (OBR) dataset to ensure robustness against variations in tumor size, position, and orientation.

Results: The proposed MRI-based model achieved an impressive accuracy of 98.76% in detecting metastatic ovarian tumors. This high level of accuracy demonstrates the model's potential as a valuable tool for early diagnosis and clinical applications.

Conclusion: The investigation confirms that the proposed 3D CNN model, leveraging MRI datasets, significantly improves the detection accuracy of metastatic ovarian tumors. This model holds promise for clinical applications, enabling timely interventions and potentially improving the life span and quality of life for patients with ovarian cancer.

CNN

MRI

Metastatic Ovarian Tumor

Ovarian Bevacizumab

Metastatic ovarian tumors (MOT) represent a significant clinical challenge in the field of women's health. These secondary tumors, originating from cancers that develop in other parts of the body, often pose intricate diagnostic dilemmas and therapeutic complexities. Early and precise detection of MOT is paramount for improving patient outcomes and tailoring effective treatment strategies [1]. In pursuit of this critical goal, this study explores the development of an advanced model for the detection of MOT, harnessing the capabilities of 3D Convolutional Neural Networks (CNNs) and innovative MRI data augmentation techniques.MOT, unlike primary ovarian cancers, arises from cancer cells that have journeyed from distant primary sites to settle in the ovaries. Common sources of these secondary tumors include breast cancer, colorectal cancer, gastric cancer, and others. Detecting MOT accurately and distinguishing them from primary ovarian malignancies is a clinical imperative, as treatment approaches and prognoses can significantly differ.

MOT remains a formidable challenge in the realm of women's health, primarily due to its elusive nature and propensity for late-stage diagnosis. Early detection of this illness is paramount for improving patient outcomes and reducing mortality rates [2]. This study delves into the development of an advanced model for the detection of metastatic ovarian tumors, utilizing the power of 3D CNNs and innovative MRI data augmentation techniques.MOT is notorious for its subtle and nonspecific symptoms, often leading to delayed diagnosis and, subsequently, limited treatment options. In the quest to identify ovarian tumors at their earliest stages, medical practitioners have traditionally relied on ultrasound imaging as a key diagnostic tool. However, emerging research and technological advancements have raised questions about the limitations of ultrasound in effectively detecting ovarian tumors, especially in their metastatic forms.

Magnetic Resonance Imaging (MRI) has emerged as a superior imaging modality for various medical applications due to its exceptional contrast, multi-dimensional visualization capabilities, and high-resolution images. In the context of ovarian cancer diagnosis, MRI offers distinct advantages over ultrasound. It provides a more detailed and comprehensive view of the pelvic region, allowing for precise localization and characterization of ovarian tumors. This study places a particular emphasis on highlighting the inherent advantages of MRI over ultrasound, shedding light on its potential to revolutionize the field of ovarian tumor detection. Furthermore, to harness the full potential of MRI in the detection of ovarian tumors, this research integrates advanced DL techniques. A 3D CNN architecture is employed, known for its prowess in image classification tasks, to analyze MRI scans. Additionally, the study introduces a novel approach to data augmentation, combining Random Cropping and Resizing with spatial deformation methods. This augmentation strategy enhances the model's capability to generalize across various tumor sizes, positions, and orientations.

The ultimate goal of this research is to present a holistic approach to ovarian tumor detection, where MRI's superior imaging capabilities and the robustness of deep learning techniques are harmoniously combined. The achieved accuracy of 97.6% serves as a testament to the potential of this approach in clinical applications, promising early detection and timely interventions in ovarian cancer cases. In doing so, this study contributes to the ongoing efforts to enhance medical diagnostics in the realm of oncology, with far-reaching implications for the well-being of patients worldwide. In summary, the main role of this effort to detect MOT is as follows,A DL model was developed for detecting metastatic ovarian tumors from MRI datasets. This investigation employed an innovative MRI data augmentation technique that combined Random Cropping and Resizing with spatial deformation methods. The work utilized Enhanced 3D Convolutional Neural Networks for image analysis in various dimensions of the input image and produced higher accuracy.

Section 2 of this paper contains the literature reviews. In section 3 of this article the methods and materials are discussed. The results and discussion are presented in Section 4.

In the multicenter study [3], advanced techniques are harnessed to enhance the detection and characterization of peritoneal metastasis in patients with epithelial ovarian cancer. The study leverages the supremacy of DL, employing CNNs to robotically extort intricate features from MRI images. Additionally, radiomics is utilized to quantify various attributes of the tumors, including texture and shape, contributing to a more comprehensive understanding of their characteristics. One notable strength of this research lies in its multicenter approach, aggregating data from multiple medical centres, which augments the dataset's diversity and size, potentially enhancing the model's applicability in clinical practice. However, variability in MRI image quality across different medical centres and imaging protocols can introduce inconsistencies and affect the model's performance. Standardization of imaging procedures is imperative to mitigate this issue. Additionally, while the multicenter study design broadens the dataset's scope, it may not fully capture the nuances of diverse patient populations. Model generalizability across different demographics and regions remains a potential challenge.

In their research, Kim and colleagues [4] presented an innovative machine learning-driven system called the Machine Learning-Based Automatic Identification of Fetal Abdominal Circumference, which aims to automatically determine fetal biometric measurements from 2D ultrasound images. This approach involves a multi-step process that incorporates a particularly premeditated CNN and a U-Net network for every stage of the analysis. These ML techniques consider the input of healthcare professionals, the underlying anatomical structures, and the distinct characteristics found in ultrasound images. The effectiveness of the approach was subsequently confirmed through the analysis of clinical data. Essentially, this method, based on ML principles, operates as a level based system that mimics the abdominal circumference reckoning procedures typically carried out by clinicians.

In their work, Brattain et al. [5] advocated the use ML methods to enhance the efficiency of ultrasound procedures in medical and clinical contexts. Ultrasound's real-time video capabilities are a notable power. The benefits of harnessing spatiotemporal data have recognized in fields like echocardiography and obstetrics, where ML methods have played a pivotal role in improving results. ML, with its capacity to uncover non-linear data characteristics, proves especially valuable in ultrasound, a domain where there are typically no straight forward or pre-defined acoustic patterns. It was designed to learn from diverse observations across different scales and modalities over time, culminating in unified machine intelligence. This intelligent system is poised to observe, guide end-users, evaluate new data, and aid in medical decision-making while also fostering the development of sustainable quantitative models. The anticipated outcome is not only a significant enhancement in clinical workflow but also an improvement in overall treatment efficacy.

Huang et al. [6] established Ultrasound Computer-aided Diagnosis with ML for the detection of fetal ultrasound planes, focusing on the facial standard plane. This approach utilizes clinical data to assess the ovarian tumor biparietal diameter and detect malformations. Efficient feature selection enhances accuracy while reducing computational complexity, addressing inconsistencies in data collection common in ultrasound CAD systems. Traditional ultrasound CAD systems typically rely on four types of common features: descriptor, textural, morphological, and model-based.

[7] In clinical practice, nuclear medicine professionals are tasked with interpreting scans related to ovarian cancers. However, this increased workload in interpreting images can elevate the risk of oversight or misdiagnosis. Consequently, there is a pressing need for an automated approach to mitigate these issues. CNNs have become known as a ML strategy in the field of medicine. CNNs, a subset of DL techniques, prove valuable for diagnostic imaging [8]. They have been extensively employed for classification, detection, segmentation, and registration in various medical imaging modalities such as MRI, Ultra sound and radiography [9,10]. Notably, CNN techniques have demonstrated effectiveness in conjunction with MRI imaging [11, 12]. For instance, Wang et al. successfully predicted ovarian cancer recurrence using MRI generated by CNNs [13], while Bogani et al [14] developed a DL model to analyze parameters predicting full cytoreduction in secondary cytoreductive surgery for recurrent ovarian cancer patients. Avanzo et al [15] primarily focused on ovarian cancer recurrence prediction using 2D CNN-based DL models. Despite a considerable body of research on DL-based medical image processing, there remains a scarcity of studies directly utilizing 3D tomographic medical imaging data [16, 17]. An overview of the literature work is presented in Table 1.

Table 1

Summary of the literature reviews
Work	Methodology	Limitations
Multicenter Study [3]	Utilizes advanced techniques and deep learning for peritoneal metastasis detection.	Variability in MRI image quality across centers can affect model performance.
	Multicenter approach enhances dataset diversity.	May not fully capture nuances of diverse patient populations.
	Radiomics for comprehensive tumor characterization.	Model generalizability across demographics and regions may be challenging.
Kim et al. [4]	Innovative ML-driven system for fetal biometric measurements.	Complex multi-step process may require specialized knowledge.
Kim et al. [4]	Utilizes both CNN and U-Net networks for analysis.	Effectiveness confirmed with clinical data, but real-world applicability may vary.
Brattain et al. [5]	Emphasizes the use of ML to enhance ultrasound efficiency.	Reliance on real-time video capabilities of ultrasound as a strength.
Huang et al. (US-CAD) [6]	Focuses on detecting standard fetal ultrasound planes.	Efficient feature selection improves accuracy but may not generalize to all cases.
	Addresses data collection inconsistencies common in ultrasound CAD systems.	Specificity to facial standard plane detection may limit broader applications.
	Architecture includes voxel-wise network and post-processing procedures.	Specific focus on tumor extraction, limited discussion on other applications.
Nuclear Medicine Professionals [7]	Highlights the need for automated approaches in interpreting ovarian cancer scans.	Elevated risk of oversight or misdiagnosis due to increased workload.
Nuclear Medicine Professionals [7]	CNNs are recognized as valuable for diagnostic imaging in medicine.	General overview of CNNs in medicine, no specific focus on ovarian cancer.
Hiratta et al [18]	Cites examples of successful CNN applications in ovarian cancer diagnosis.	Lack of studies directly utilizing 3D tomographic medical imaging data.

Ovarian cancer remains a formidable challenge in the realm of oncology, with metastatic spread being a critical determinant of patient prognosis and treatment strategies. The accurate and early detection of metastatic lesions in ovarian tissue is imperative for effective clinical decision-making. In recent years, the advent of DL algorithms, particularly 3D CNN, perform excellently in the field of medical and health image analysis. This research leverages the power of 3D CNNs to address the vital task of metastatic ovarian tumor detection. By harnessing advanced computational methods and meticulously curated histopathological datasets, we aim to enhance the precision and reliability of MOT diagnosis. The proposed Enhanced 3D CNN is as shown in Figure.1A comprehensive account of our methods and materials, dataset, pre-processing procedures, architecture of the Enhanced 3D CNN, training methodology, and evaluation metrics employed to assess the performance of our model are presented as follows.

A. Dataset

The Ovarian Bevacizumab Response (OBR) dataset [19] comprises around 288 both benign and metastatic cases with hematoxylin and eosin stained (basic reason for tumor) whole slides, accompanied by clinical details from 78 patients. These slides were sourced from the tissue bank of the Tri-Service General Hospital in Taipei, Taiwan. Acquisition of these MR images was done using Leica AT2 scanner equipped with a 20x objective lens. The ovarian cancer slides have an average dimension of 54342x41048 pixels and measure approximately 27.34x20.66mm. The dataset were sourced from https://www.cancerimagingarchive.net/collection/ovarian-bevacizumab-response/

B. Data Augmentation

Data augmentation was instrumental in diversifying and expanding the MRI image dataset for the critical task of detecting metastatic ovarian tumors. This augmentation process involved a combination of techniques, including random cropping, resizing, and spatial deformation. Random cropping was employed to select random regions of the original images, followed by resizing to ensure uniformity. However, these techniques alone had limitations in capturing the complexity of real-world tumor appearance variations, especially for tumors with intricate morphologies and variations in imaging conditions. Recognizing this limitation, spatial deformation was introduced to the augmentation pipeline. Spatial deformation added realistic distortions, mimicking the nuanced variability seen in clinical scans. The decision to combine these techniques was justified by the necessity to strike a balance between dataset consistency and replicating genuine tumor appearance variations. This approach ensured that our deep learning model learned not only from a diverse range of images but also from augmented data that closely resembled the clinical complexity of metastatic ovarian tumors. Consequently, the model's ability to accurately detect tumors and generalize to unseen data was greatly enhanced.

The number of images generated through data augmentation varies due to the factors, together with the chosen augmentation parameters and the extent of augmentation applied to each original image. In one approach, random cropping and resizing are performed, resulting in approximately 5 augmented images produced from every original MR. Additionally, when applying spatial deformation, an average of 5 additional augmented images are produced from every original MR image. A sample of augmented images is shown in Figure.2.We started with an original dataset containing 632 MRI images, the full amount number of augmented MR images can be calculated as follows: For random cropping and resizing, it yields 3160 augmented images (632 original images multiplied by 5 augmented images each). For spatial deformation, a similar result of 3160 augmented images is obtained. When both augmentation techniques are combined, the total number of augmented images sums up to 6320.Table 2 provides the image samples used for this investigation.

Table 2

Image samples obtained from augmentation
Augmentation Technique	Number of Augmented Images per Original Image	Total Augmented Images
Random Cropping and Resizing	5	3160
Spatial Deformation	5	3160
Combined (Both Techniques)	N/A	6320

C. Enhanced 3 Dimensional CNN

A specialized deep learning model known as a 3 Dimensional CNN is well-suited for the task of identifying metastatic ovarian tumors in MRI images. In contrast to 2D CNNs, which analyze two-dimensional images, 3D CNNs are tailored to handle the huge amount of MRI data, enabling them to incarcerate spatial relationships in all 3 dimensions. These networks employ a sequence of convolutional layers to extract complex features, recognizing patterns and textures throughout MRI volumes. Data augmentation techniques are employed to enhance the model's adaptability, introducing variations in tumor attributes, including size, position, and orientation, within the training dataset. Through training and validation processes, the model learns and fine-tunes its parameters, while feature maps depict essential learned characteristics. Additionally, the model can perform segmentation, delineating tumor boundaries and regions of interest within the MRI volume. Ultimately, this model serves as a potent tool for early detection and improved patient outcomes in cases of ovarian cancer, thanks to its capability to process 3D data and capture crucial spatial relationships vital for tumor identification and localization. The Enhanced 3D CNN architecture was integrated with residual connections. It includes two specialized residual sectors, with one of them featuring an additional layer in its skip connection. The detailed representation of this architecture is referred in Table 3.

Table 3

Architecture Details of Enhanced 3D CNN
3D CNN Layer	Output Size
Input Layer	(128, 128, 128, 1)
Zero Padding	(132, 132, 132, 1)
3D Convolutional	(64 ,64, 64, 1)
Batch Normalization	(64, 64, 64, 1)
Activation	(64, 64, 64, 1)
3D max pooling	(32, 32, 32, 64)
Residual Layer 1 & 2	(32, 32, 32, 256)
Residual Layer 3 & 4	(16, 16, 16, 512)
Residual Layer 5 & 6	(8, 8, 8, 1024)
Residual Layer 7 & 8	(4, 4, 4, 2048)
3D Average pooling	(2, 2, 2, 2048)
Fully connected Layer	1

The convolution process is employed on a three-dimensional input volume, MRI. The mathematical expression for an individual convolution operation at a particular position within the input volume is as follows: For each position (a, b, c) in the output feature map is the sum of products of elements in the input volume (J) and the corresponding elements in the 3D convolution kernel (L).A stride (s) was applied to determine the step size and padding (p) was included by adding zeros around the input volume.

$$\:\left(\:J\:\times\:L\right)\left(a,b,c\right)={\sum\:}_{i=0}^{n-1}{\sum\:}_{i=0}^{m-1}{\sum\:}_{i=0}^{s-1}(a.s+x-p.b.s+y-p,c.s+L-a.L\left(x,y,z\right))$$

Following the convolution operation, a non-linear activation function like the Rectified Linear Unit (ReLU) is employed on an element-by-element basis. The ReLU function (RL) is represented as:

RL(x) = max (0, x) (2)

This equation substitutes any negative values in the feature maps with zeros, facilitating the network’s ability to discern intricate patterns. Pooling layers serve to decrease the spatial dimensions of feature maps while preserving vital information. Max Pooling (M) selects the maximum value within a specified pooling window, and its formula is expressed as:

M (a,b,c) = max l,m,n (J(a.s + l,b.s + m.s + m,c.s + L)) (3)

s determines the stride, controlling how much the pooling window shifts in each dimension. (a,b,c) denotes the position in the resulting pooled output. A fully connected layer links the results from the convolutional and pooling layers to form a vector that has been flattened. The mathematical representation of a dense layer's operation is given by:

FL (i) = Wt + m (4)

I is the input vector, Wt and m are weight matrix and bias vector respectively. In the final layer of a classification network, SoftMax activation is applied to transform the network’s initial scores into probabilities for each class. The SoftMax function (S) can be defined as follows: ti signifies the score for class i.CL represents the total number of classes.

S= $\:\frac{{e}^{ti}}{\sum\:_{j=1}^{CL}{e}^{tj}}$ (5)

The generation of feature maps in this architecture primarily relies on 3-dimensional Convolution layers, effectively extracting critical image features from the volumetric MRI data.The enhanced 3-dimensional CNNs processed data in batches for higher classification accuracy. With batch processing, the convolution operation is performed on a batch of input volumes, and the result is computed for each volume in the batch. A bias term (m) is added to the output of the convolution operation for each filter. This bias term is an additional learnable parameter that affects the final output.

$$\:\left(\:J\:\times\:L+m\right)\left(a,b,c\right)={\sum\:}_{i=0}^{n-1}{\sum\:}_{i=0}^{m-1}{\sum\:}_{i=0}^{s-1}(a.s+x-p.b.s+y-p,c.s+L-a.L\left(x,y,z\right))+m$$

To determine the dimensions of the output feature map, width (Ow), height (Oh), and depth (Od) of the output are calculated based on the dimensions of the input volume, the convolution kernel, stride (S), width of the kernel (Wk), height of the kernel (Hk), depth of the kernel (Dk) and padding (P). These formulas ensure that the output feature map size is appropriately adjusted.

Ow = $\:\:\:\frac{\:W-Wk+2.P}{8}\:+1\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:$ (7)

Oh = $\:\:\:\:\:\frac{W-Hk+2.P}{8}\:+1\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:$ (8)

Od = $\:\:\:\:\frac{W-Dk+2.P}{8}\:+1\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:$ (9)

In essence, the convolution operation in an Enhanced 3D CNN involves sliding a 3D kernel over the input volume, performing element-wise multiplications, accumulating results, and adding a bias term when necessary. This process is repeated across the entire input volume to produce the final feature map. The image collections were divided into three distinct groups: 75 per cent of the dataset was used for training set, for validation 10 per cent, and 15 per cent for testing set. To enhance the training sets, augmentation techniques such as random rotations and Spatio deformation were applied. The validation set played a pivotal role in the selection of various model hyper parameters, including aspects like learning rate, architecture of the DL network, and decay steps. This process allowed us to identify and establish the best-performing 3D CNN model. Subsequently, the test subset was employed to evaluate the performance of the proposed model. The networks underwent a training process spanning 100 epochs, employing a batch size of 4.Adam optimizer along with learning rate of 0.00001 was used. Training concluded when the validation accuracy reached a point of stabilization and ceased to increase. To create the proposed model, we utilized Python programming language in conjunction with TensorFlow version 2.9.1 and the Keras framework version 2.9.0 tailored for neural networks. The DL processes were conducted on a computer device equipped with i7 core CPU running at 3.60 GHz, 64 GB of RAM, and RTX 3090 GPU.

In our study, we conducted an experiment utilizing a 3D CNN to analyze MRI datasets for the detection of metastatic ovarian tumors. Our model yielded impressive results, with 98.76% of accuracy, 98.11% of recall, 98.19 of precision, F1 score of 98.31, error rate of 1.29% .These metrics are crucial in evaluating the performance of a ML model, and they provide a comprehensive assessment of its capabilities. The overall results are shown in Fig. 3. The key metrics used in our assessment are calculated as follows: Correctly classified samples out of the total samples are termed as accuracy of a model and are deliberated as in Eq. (10).

Accuracy = (True-Positive + True-Negative) / (True-Positive + True-Negative + False-Positive + False-Negative) (10)

Where True-Positives represents correctly predicted positive cases, True-Negatives represents correctly predicted negative cases, False-Positives represents incorrectly predicted positive cases, and False-Negatives represents incorrectly predicted negative cases.

Recall is the model’s ability to correctly identify positive cases and is calculated as in Eq. (11). It indicates the model's ability to detect metastatic ovarian tumors accurately.

Recall = True-Positive / (True-Positive + False-Negative) (11)

Negatively classified samples of the model are termed as specificity and are calculated as in Eq. (12). It reflects how well the model can distinguish non-tumor cases.

Precision = True-Negative / (True-Negative + False-Positive) (12)

The harmonic mean of recall and precision is F1 score and premeditated as in Eq. (13)

F1 Score = 2 × (Precision × Recall) / (Precision + Recall) (13)

It balances the trade-off between precision and recall, providing a comprehensive performance metric. The error rate is the percentage of misclassified cases and is calculated as in Eq. (14). It quantifies the overall model's error.

Error Rate = (False-Positive + False-Negative) / (True-Positive + True-Negative + False-Positive + False-Negative) (14)

Using augmentation techniques to enhance the input image data is a positive aspect of our approach. Data augmentation helps to diversify the dataset, reducing overfitting and improving the model's generalization ability. By introducing variations in the data, the model becomes more robust to different image conditions and angles, ultimately enhancing its performance in detecting metastatic ovarian tumors.Our study demonstrates the effectiveness of a 3D CNN model in detecting metastatic ovarian tumors, as evidenced by the high accuracy, recall, precision, F1 score, and low error rate. Additionally, the utilization of data augmentation techniques further strengthens the model's robustness and reliability, making it a promising tool for early diagnosis and treatment of ovarian cancer. Table 4 shows the overall results obtained for the proposed approach.

Table 4

Results of Enhanced 3DCNN in identifying MOT
K fold	Accuracy (%)	Recall (%)	Precision ( %)	F1 Score (%)
3	97.24	97.21	96.14	96.03
4	97.31	97.02	97.43	97.42
5	98.01	97.42	97.69	97.96
10	98.76	98.11	98.19	98.31

The proposed approach, which employs an Enhanced 3D CNN for the detection of metastatic ovarian tumors in MRI datasets, has demonstrated remarkable performance metrics. It outperforms several existing approaches in various critical aspects. The performance metrics of the 3D Convolutional Neural Network (3D CNN) model for detecting metastatic ovarian tumors from MRI scans were assessed across various K values, demonstrating the model's efficacy and the influence of K on its results. For K = 3 (Fig. 4), the model attained an accuracy of 97.24%, a recall of 97.21%, a precision of 96.14%, and an F1 score of 96.03%. When K was set to 4 (Fig. 5), the metrics slightly improved, with the accuracy reaching 97.31%, recall at 97.02%, precision at 97.43%, and an F1 score at 97.42%. At K = 5 (Fig. 6), the model's accuracy further increased to 98.01%, with a recall of 97.42%, precision of 97.69%, and an F1 score of 97.96%, reflecting a well-balanced performance in identifying and correctly predicting positive cases. The model's best performance was noted at K = 10 (Fig. 7), with an accuracy of 98.76%, recall of 98.11%, precision of 98.19%, and an F1 score of 98.31%. These findings indicate that the model, especially at K = 10, maintains a strong balance between correctly identifying true positives and minimizing false positives, proving to be an effective tool for enhancing the detection of metastatic ovarian tumors using MRI-enhanced imaging.

Firstly, our approach showcases superior memory utilization. By utilizing 3D CNN architecture, we efficiently process volumetric MRI data while minimizing memory consumption. This is especially crucial for large medical imaging datasets, where memory-efficient models can significantly reduce hardware requirements and facilitate real-world deployment. Secondly, our model excels in terms of trainable parameters. It also allows for a more compact representation of the underlying features, resulting in fewer parameters to train compared to some other complex architectures. Fewer trainable parameters not only make the training process faster but also reduce the risk of overfitting, leading to a more robust model.

Table 5

Accuracy wise Comparison of Proposed Approach with Existing Approaches
S.No	Model	Accuracy	F1 Score
1	Clinical model	89.32%	89.03%
2	OCDAC Net	92.01%	91.91%
3	ML CNN	94.64%	94.36%
4	2D CNN	91.09%	91.01%
5	Proposed Model	98.6%	98.31%

Furthermore, our approach demonstrates competitive execution time. The efficiency of 3D CNNs in processing volumetric data, combined with our careful consideration of data augmentation techniques, ensures that the model can make predictions swiftly. This rapid execution time is crucial in clinical settings, where timely diagnosis and treatment decisions are paramount. The proposed approach is compared with few of the existing approaches that used DL models to detect the cancer. The accuracy wise comparison is shown in Table 5 and Fig. 8.

The proposed approach stands out in terms of memory utilization, trainable parameters, and execution time efficiency as shown in Table 6, making it a promising solution for the early detection of metastatic ovarian tumors in MRI datasets. Its superior performance metrics further reinforce its potential as advancement over existing methods, with the potential to significantly impact the field of medical image analysis and diagnosis.

Table 6

Metrics Obtained for Proposed Approach
Metrics	Metrics Obtained
Memory	36 MB
Execution time	0.7562ms
Time to predict	10.43ms
Accuracy	98.6%

To reduce data size and enhance resistance to misalignment, 3D Pooling layers performed down sampling on the resulting feature map. Inspired by ResNet, the "Residual" sector facilitates layer combination, effectively mitigating the issue of gradient vanishing commonly encountered in traditional networks. Neurons within a layer maintain connections with related neurons in the preceding layer, fostering information flow. Additionally, this network incorporates batch normalization, sigmoid layers, and utilizes the rectified linear unit function (ReLU) to enhance its overall performance. The fusion of scale-invariant networks with this sophisticated 3D CNN architecture creates a powerful tool for metastatic ovarian tumor detection in MRI scans. This combined approach ensures adaptability to tumor sizes and shapes within the volumetric data, offering precise and accurate identification of tumors regardless of their dimensions. The Residual connections enable the efficient integration of layers, addressing gradient-related challenges and enhancing the network's ability to capture nuanced features. This comprehensive architecture holds significant potential in improving diagnostic accuracy and clinical outcomes in the field of oncology.

Enhanced 3D CNN utilized in the proposed approach offered several advantageous features when it comes to identifying images of MOT MRI dataset. The proposed 3D CNN excelled at capturing spatial and temporal dependencies in volumetric data, by making it well-suited for MOT imaging tasks. Unlike traditional 2D CNNs, which treat each image slice independently, the proposed Enhanced 3D CNN considered the entire 3D volume, preserving critical 3D structural information in MRI data. This holistic view enabled the model to capture subtle patterns and correlations that might be missed by 2D counterparts. The proposed 3D CNN effectively reduced the risk of false positives and false negatives in MOT image analysis. They considered the context of neighboring slices, aiding in distinguishing genuine abnormalities from artifacts or noise. Additionally, it is capable of involuntarily extracting hierarchical input features, relieving the necessity for labor-intensive feature engineering, which is often challenging in complex medical images.

In summary, this research presents an efficient method for identifying metastatic ovarian tumors within MRI datasets, employing a 3D CNN. The model achieved exceptional results with 98.61% accuracy, sensitivity of 98.7%, specificity of 98%, and F1 score of 98.31%. Additionally, our model demonstrated a low error rate of just 1.29% and a recall rate of 0.97, illustrating its reliability and its ability to minimize both false positives and false negatives. These findings underscore the substantial potential of our Enhanced 3D CNN approach in clinical applications, where early and accurate detection of metastatic ovarian tumors is of paramount importance. Furthermore, by incorporating data augmentation techniques to enhance the input images, we bolstered the model's resilience and its ability to generalize effectively. This study symbolize a notable encroachment in medical image analysis, offering a promising tool for enhancing patient outcomes through early diagnosis and intervention in ovarian cancer cases.

2D CNN − 2D Convolutional Neural Networks

3D CNN − 3D Convolutional Neural Networks

CAD - Computer-Aided Diagnosis

DL - Deep Learning

ML - Machine Learning

MOT - Metastatic Ovarian Tumors

MRI - Magnetic Resonance Imaging

OBR - Ovarian Bevacizumab Response

RL - ReLU (Rectified Linear Unit) function

MB – Megabyte

Acknowledgements

The authors thank VIT for providing VIT SEED GRANT (RGEMS) – Sanctioned Order No. SPL/SG20230148 for carrying out this research work.

Author Contributions: P.R and M. T.; methodology, P.R., M. T., G.A software, M.T.; validation, G.A., M. T. and J.J.; formal analysis, G.A.; investigation, G.A.; resources, G.A.; data curation, G.A; writing—original draft preparation, P.R.; writing—review and editing, J.J.; supervision, J.J.; project administration, J.J. All authors have read and agreed to the published version of the manuscript.

Availability of data and materials

The data that support the findings of this study are openly available at https://www.cancerimagingarchive.net/collection/ovarian-bevacizumab-response/

Funding

Vellore Institute of Technology, Vellore, Tamil Nadu, India has provided financial support for the project.

Ethics approval and consent to participate

Not Applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Cheng F, Wang Y. Research and application of 3D visualization and Internet of Things technology in urban land use efficiency management. Displays. 2021;69:102050.
Dondi F, Albano D, Bertagna F, Giubbini R. [18F] FDG PET/CT and CA-125 in the evaluation of ovarian cancer relapse or persistence: is there any correlation? Nuclear Med Rev. 2022; 25(2):78–84. Ovarian Cancer. - Cancer Stat Facts [Internet] [Accessed 2023 Jan 2]. https://seer.cancer.gov/statfacts/html/ovary.html
Wei M, Zhang Y, Ding C, Jia J, Xu H, Dai Y, Feng G, Qin C, Bai G, Chen S, Wang H. Associating Peritoneal Metastasis with T2-Weighted MRI Images in Epithelial Ovarian Cancer Using Deep Learning and Radiomics: A Multicenter Study. J Magn Reson Imaging, 2023.
Kim B, Kim KC, Park Y, Kwon J-Y, Jang J, Seo JK. Oct., Machine learning-based automatic identification of fetal abdominal circumference from ultrasound images. Physiological Measures, 39(10), 2018, Art. 105007.
Brattain LJ, Telfer BA, Dhyani M, Grajo JR, Samir AE. Machine learning for medical ultrasound: Status, methods, and future opportunities. Abdom Radiol. Feb. 2018;43(4):786–99.
Huang Q, Zhang F, Li X. Machine learning in ultrasound computer aided diagnostic systems: A survey. Bio Med Res Int., 2018.
Xu Y, Yang J, Zhang Z, Zhang G. MRI for discriminating metastatic ovarian tumors from primary epithelial ovarian cancers. J Ovarian Res. 2015;8:61. 10.1186/s13048-015-0188-5.
Saba L, Biswas M, Kuppili V, Godia EC, Suri HS, Edla DR, et al. The present and future of deep learning in radiology. Eur J Radiol. 2019;114:14–24.
Satoh Y, Imokawa T, Fujioka T, Mori M, Yamaga E, Takahashi K, et al. Deep learning for image classification in dedicated breast positron emission tomography. Ann Nucl Med. 2022;36(4):401–10. https://doi.org/10.1007/s12149-022-01719-7.
Sharma G, Prabha C. A systematic review for detecting cancer using machine learning techniques, In: AIP Conference Proceedings, AIP Publishing LLC; 2022. p. 40007.
Akazawa M, Hashimoto K. Artificial intelligence in gynecologic cancers: Current status and future challenges–A systematic review. Artif Intell Med. 2021;120:102164.
Saba T. Recent advancement in cancer detection using machine learning: Systematic survey of decades, comparisons and challenges. J Infect Public Health. 2020;13(9):1274–89.
Rajesh P, Kavitha R. Elderly people activity monitoring with involved binary sensors and Deep Convolution Neural Network, (2022), Neural Computing and Applications, https://doi.org/10.1007/s00521-022-07268-4
Bogani G, Rossetti D, Ditto A, Martinelli F, Chiappa V, Mosca L et al. Artificial intelligence weights the importance of factors predicting complete cytoreduction at secondary cytoreductive surgery for recurrent ovarian cancer. J Gynecologic Oncol, 2018; 29(5).
Avanzo M, Porzio M, Lorenzon L, Milan L, Sghedoni R, Russo G, et al. Artificial intelligence applications in medical imaging: A review of the medical physics research in Italy. Phys Med. 2021;83:221–41.
Trivizakis E, Manikis GC, Nikiforaki K, Drevelegas K, Constantinides M, Drevelegas A, et al. Extending2-D convolutional neural networks to 3-D for advancing deep learning cancer classification with application to MRI liver tumor differentiation. IEEE J biomedical health Inf. 2018;23(3):923–30.
Salehi SSM, Hashemi SR, Velasco-Annis C, Ouaalam A, Estroff JA, Erdogmus D, Wareld SK, Gholipour A. Real-time automatic fetal brain extraction in fetal MRI by deep learning,in Proceedings of IEEE 15th Int. Symp. Biomed. Imag, (ISBI), Apr. 2018, pp. 720–724.
Hirata K, Sugimori H, Fujima N, Toyonaga T, Kudo K. Artificial intelligence for nuclear medicine in oncology. Ann Nucl Med. 2022;36(2):123–32. https://doi.org/10.1007/s12149-021-01693-6.
Wang C-W, Chang C-C, Lo S-C, Lin Y-J, Liou Y-A, Hsu P-C, Lee Y-C, Chao T-K. A dataset of histopathological whole slide images for classification of Treatment effectiveness to ovarian cancer (Ovarian Bevacizumab Response) (Version 2) [Data set]. Cancer Imaging Archive. 2021. https://doi.org/10.7937/TCIA.985G-EY35.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

MRI-Enhanced Metastatic Ovarian Tumor Detection: Leveraging Enhanced 3D CNN and Data Augmentation for Exceptional Accuracy

Status:

Version 1

Abstract

Figures

1. INTRODUCTION

2. RELATED WORKS

3. Methods and Materials

4. Results and Discussion

5. Conclusion

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1