Breast Cancer Prediction Using Machine Learning: A YOLOv8 Approach

doi:10.21203/rs.3.rs-4976943/v1

Breast cancer is among the major concerns in global health, and its management starts with early diagnosis. This article presents an advanced machine learning approach with a deep learning YOLO algorithm (You Only Look Once). YOLOv8 is the definitive version of the YOLO deep learning algorithm. The breast cancer detection YOLOv8 model is based on ultrasound images. In the given case, deep learning techniques are being ended with to give detection high precision, speed, and performance. This paper presents an application of a deep learning algorithm, YOLOv8, in real-time breast cancer detection using ultrasound imaging. In comparison, this model represented higher accuracy and recall than both ResNet50 and VGG16, thereby representing good potential for its integration into clinical settings. Our model showed results of 93% accuracy and 92% recall, which exceeds the results of ResNet50 and VGG16 by 6% and 10%, respectively. Finally, we have described how the integration of this system will be implemented on a clinical level in a real-time web-based interface, closing our work and showing future work at the clinical level how this research may be a source of such advancements in the early detection of breast cancer.

Breast cancer

Machine Learning

Deep Learning

YOLO deep learning

YOLOv8

Ultrasound Imaging

Cancer Detection

Web-based Interface

Early Diagnosis.

Breast cancer is the most frequently diagnosed cancer in women, and thus, early diagnosis becomes even more imperative for enhancing survival rates. However, traditional techniques often go awry in women with dense breast tissue, creating an urgent need to devise something more innovative [1]. Early diagnosis is essential to improving patient outcomes, with imaging techniques crucial in detecting tumors at an earlier, more treatable stage. Mammography has long been the gold standard for breast cancer screening; however, it is less effective for women with dense breast tissue, often leading to false negatives [2]. Ultrasound imaging has emerged as a valuable complementary modality, especially in detecting tumors that are not visible on mammograms [3].

Breast cancer is a complex and heterogeneous disease characterized by the uncontrolled growth of abnormal cells in the breast tissue. While it primarily affects women, breast cancer can also occur in men, albeit at a much lower frequency. The disease can manifest in various forms, ranging from non-invasive tumors confined within the ducts or lobules of the breast (ductal carcinoma in situ, or DCIS) to invasive tumors that penetrate surrounding tissue and may spread to distant organs (invasive ductal carcinoma, or IDC) as shown in Fig. 1.

Despite its advantages, ultrasound interpretation is highly dependent on the expertise of radiologists, which can introduce subjectivity and variability. Automated methods, particularly those leveraging machine learning, offer a potential solution by standardizing and improving the accuracy of diagnosis. Convolutional neural networks (CNNs) have demonstrated success in image analysis tasks, and recent advances in object detection models, such as the YOLO (You Only Look Once) family, have shown promise in medical imaging applications [4].

This paper presents an application of the YOLOv8 model to detect breast cancer from ultrasound images. By using a deep learning method, we aim to improve detection accuracy, reduce false positives and false negatives, and enable real-time diagnostic capabilities. The study also explores the integration of this model into a web-based system that facilitates remote access and usability for clinicians in various healthcare settings.

Novel contributions of this work are the integration of the YOLOv8 model within a clinical framework that enables remote access and real-time diagnostics.

Many machine learning techniques have been intensely studied for breast cancer recognition. Such traditional methods include SVM and k-NN, which work effectively for their handcrafted features but can handle little meaningfulness of medical images [5]. When analyzing medical images, deep learning and more specifically the application of CNNs have become immensely powerful tools because of their ability to automatically learn and extract the features from raw images.

Initial work with CNN in the identification of breast cancer included such models as VGG16 and ResNet, which continue showing elevated levels of accuracy in multiple classification tasks [6]. However, real-time performance is among the biggest drawbacks and hinders the application of these in practice.

These tasks were computationally more efficiently solved on the fly with the YOLO architecture, relying only on the ability to predict multiple classes simultaneously within a single pass through the network [7].

While some studies have utilized YOLO with medical imaging, significantly fewer have directly adapted this framework to facilitate the investigation of breast cancer detection through ultrasound images, which is why the current research is so innovative. This research fills in that gap by adapting YOLOv8 to this critical task, making it possible to harness real-time processing and achieve better detection accuracy.

3.1. Data Collection and Preprocessing

The dataset comprised 1,000 images of breast lesions from ultrasound, which were collected from open databases and clinical archives. The dataset contains breast ultrasound images of 600 female patients; age is not less than 25 and not more than 75. Image collection was done in 2018. The size is 780 images in PNG format, of an average size of 500x500 pixels [8]. The images were preprocessed to a standardized resolution of 64x64 pixels to ensure uniformity across the dataset. We fine-tuned the YOLOv8 model on this dataset for 50 epochs using a batch size of 64.

This dataset had balanced quantities of both benign and malignant cases. Variables of different shapes, sizes, and textures represent benign and malignant lesions. It involves a range of preprocessing techniques that include noise reduction; contrast improvement through median filtering; and resizing the images to standardize them, thus increasing the performance level of the model. It applies data augmentation techniques of rotations, flip, and zoom on the dataset to add more variability applied into the training set to prevent overfitting. Figure 2 provides a sample of data, and Table 1 represents the summary of the used dataset in terms of how many images for each class were applied to the augmentation methods [9],[10].

Table 1

A summary of the used dataset
Lesion Type	Number of Images	Data Augmentation Techniques
Benign	500	Rotation, Flip, Zoom
Malignant	500	Rotation, Flip, Contrast Adjustment

3.2. YOLOv8 Model Architecture

The YOLOv8 represents a significant advancement in the YOLO series by prioritizing improvements in both accuracy and speed. It introduces anchor-free detection and a more efficient backbone, allowing it to process images in real-time with minimal computational overhead. The model's architecture consists of a feature extraction network followed by detection layers that predict the occurrence and place of objects within the image.

The network was fine-tuned by using transfer learning in the world of breast cancer detection by working on a pre-trained model on the COCO dataset when detecting Common Objects in Context [10].

It optimizes the hyper-parameters during training for learning rate, batch size, and epochs. Early stopping avoided overfitting of the model while saving checkpoints based on validation performance.

The architecture of the YOLOv8 model is illustrated in Fig. 3 and shows the flow of information from input to prediction.

The training process involved optimizing the model's hyperparameters, containing the batch size, learning rate, and number of epochs. Early stopping was employed to prevent overfitting, and model checkpoints were saved based on validation loss [11].

3.3. Web-based Accessibility: Convenience at Your Fingertips

One of the defining characteristics of this project lies in its web-based architecture. This strategic choice prioritizes accessibility and user convenience. The platform will be accessible through any standard web browser, without the need for users to download and install additional software. This approach caters to a wide range of users, from healthcare professionals in well-equipped medical facilities to individuals with limited access to specialized technology.

Uploading of the ultrasound images for analysis will be very easy, front-ended by a user-friendly interface with least technical knowledge required. The flowchart of the process for ultrasound image classification as benign or malignant is presented in Fig. 4.

This design consideration ensures that the platform's benefits can be widely disseminated, fostering inclusivity and democratizing access to this sophisticated diagnostic tool as shown in Fig. 5.A and Fig. 5.B.

Figure 4.B demonstrates that patients can easily upload their medical scans with the "Upload your scan" button. The predict button will be enabled after the image passes verification of authenticity and quality. predict button employs advanced algorithms to analyze uploaded scans, predict cancer types and characteristics, and output the prediction results.

3.4. Training of the neural network model

To analyze the neural network training, we use the Neural Network Training Diagram Fig. 6.

The diagram is a graphical representation of training a neural network model. Along the x-axis is represented the number of training epochs (One complete pass through the entire training dataset), and on the y-axis are the loss and accuracy metrics. The subplots are four: two for loss functions and two for accuracy metrics, classified in top-1 and top-5.

Loss Functions

Train/Loss: This graph shows how the model loss changes with time. When dropping, the model is learning and its performance is improving when assessed on train data.
Val/Loss: This is the graph that describes the loss of the model on the validation dataset. It should be taken under consideration because there is a risk of getting into the zone of overfitting while this number goes up. This actually happens when your model becomes overfit and increasingly specialized on the trained data, thus performing weakly on new data. It is indicated by increasing validation loss while the training loss keeps going down.

Accuracy Metrics

Metrics/Accuracy_top1: This curve shows the percentage of correct predictions where the top-estimated class is one of the ground truth labels. A higher curve indicates better overall performance.
Metrics/Accuracy_top5: This curve shows the percentage of correct predictions where the correct class is among the five most probable predicted classes. In case of multi-class classification, it is a frequently used measure since in practice it allows considering lower-ranked predicted classes.

Observations

Training Loss: The training loss always decreases, which means the model is learning well. Validation Loss:
The validation loss is first decreasing and then saturated, implying that the model could have started to overfit.
Accuracy: Both top-1 and top-5 accuracies are increasing, leading to the conclusion that the model is doing better than before at the classification of examples.
Smoothness: The "smooth" line in the validation loss plot may be its smoothed version of the original curve of loss, which can help obtain the trends clearly.

In general, the plots relate to normal training processes where models tend to improve relative to their performance on training and validation data, which comes at the risk of overfitting and therefore should continue to be watched [10].

3.5. Evaluation Metrics

The performance of the YOLOv8 model was benchmarked using, precision, accuracy, F1-score, and recall. In addition, confusion matrices were generated to provide insights into the model's classification performance. The evaluation was conducted on a test set that was not used during training or validation to ensure that the results were representative of the model's ability to generalize to new data [12].

Evaluation Metrics descriptions:

Precision: Number of correct positive predictions by the model.
Recall: The recall presents how much the true positive cases have been shown by the model.
F1-score: It is a balanced measure since the harmonic mean of precision and recall has to be used.
Accuracy: This is the total percentage of the correct predictions, both positive and negative.

Table 2 presents the model results and performance metrics for the YOLOv8 model compared to other popular CNN models used in breast cancer detection.

Table 2

The model results and performance metrics
Model	Precision	Recall	F1-Score	Accuracy
YOLOv8	0.93	0.92	0.92	90%
ResNet50	0.87	0.85	0.86	88%
VGG16	0.83	0.81	0.82	85%

The YOLOv8 model achieved a precision of 93%, recall of 92%, and a general accuracy of 90% in detection of breast lesions. These results indicate a significant improvement over traditional CNN models like ResNet50 and VGG16, as shown in Table 2 and Fig. 7. The high precision reflects the model's ability to correctly identify malignant lesions, while the high recall demonstrates its capability to detect most of the actual cancerous cases [13].

Results Interpretation:

High Precision: The high precision of YOLOv8 means that, in case of predicting a lesion to be malignant, the probability of this being the right prediction is quite great. This is of great importance in medical applications for avoiding false positives.
High Recall: A high recall of YOLOv8 indicates that nearly all true malignant lesions could be detected, therefore decreasing the risk of a false negative result.
Overall Accuracy: The overall accuracy for YOLOv8 was 90%, which means it did well in detecting both benign and malignant lesions.

A confusion matrix (Fig. 8) was generated to visualize the model's classification performance, providing a clear picture of how the model differentiates between malignant and benign lesions.

Here's a breakdown of the matrix:

True Positive (TP): 500 benign lesions correctly classified as benign.
False Positive (FP): 1 benign lesion incorrectly classified as malignant.
True Negative (TN): 399 malignant lesions correctly classified as malignant.
False Negative (FN): 0 malignant lesions incorrectly classified as benign.

From these, we can derive the following:

Accuracy: (TP + TN) / (TP + FP + TN + FN) = (500 + 399) / (500 + 1 + 399 + 0) ≈ 99.75%
Precision: TP / (TP + FP) = 500 / (500 + 1) ≈ 99.80%
Recall: TP / (TP + FN) = 500 / (500 + 0) = 100%
F1-Score: 2 * (Precision * Recall) / (Precision + Recall) = 2 * (0.9980 * 1) / (0.9980 + 1) ≈ 99.90%

Interpretation:

High Accuracy: The model was 99.75% accurate, so it was correct in most of the cases.
High Precision: The precision is 99.80%, so every time it predicts a lesion as benign, it is highly likely to be correct.
Perfect Recall: The 100% recall signifies that the model was able to identify all of the malignant lesions, which kind of strikes at the core of the medical application.
High F1-Score: An F1-score of 99.90% depicts great overall performance, balancing precision and recall.

Another important observation from this study was the real-time processing of images by the model, making it quite feasible for integration into clinical workflows. The YOLOv8 model can analyze ultrasound images in under twenty milliseconds per image, allowing faster decision-making and reducing workload for radiologists [13].

This technique, however, had some limitations, especially in the recognition of smaller or poorly contrasted lesions. Most of these cases often resulted in false negatives where it failed to detect a malignant lesion. Future work should target enhancing the sensitivity of the model to such challenging cases by incorporating multi-scale feature extraction techniques or combining ultrasound data with other imaging modalities like MRI [14].

In the paper, we propose a new application of the YOLOv8 model for breast cancer detection in ultrasound images. This model has high accuracy and is processed in real time, which may open promising avenues for application in the clinic. The potential to automate detection will facilitate fast and accurate diagnoses by radiologists and increase the chances of better patient outcomes.

According to our results, YOLOv8 one of the most effective models in detecting breast lesions due to its results, which are better than those of other traditional CNN architectures like ResNet50 and VGG16. This is a very useful tool within medical imaging, with high precision, recall, and accuracy.

In the future, it will be extended to diverse types of breast lesions. Moreover, there is a necessity to integrate this model into a cloud-based system so that remote diagnosis can also be achieved. Further studies are underway to get over the limitations shown by this study in detecting small or poorly contrasted lesions.

Author Contribution

I was the primary investigator, responsible for conceptualizing the study, designing experiments, and analyzing the data. I wrote the initial draft of the manuscript and oversaw the review process.

Acknowledgement

Special thanks to Engineers Khaled Ahmed Foad and Omar Ahmed Hassan for their effective efforts in implementing the system's web application.

M. Masud, et al., “Pre-Trained Convolutional Neural Networks for Breast Cancer Detection Using Ultrasound Images,” ACM Transactions on Internet Technology, vol. 21, no. 4, pp. 1–17, 2021.
D. Thigpen, A. Kappler, and R. Brem, “The Role of Ultrasound in Screening Dense Breasts—A Review of the Literature and Practical Solutions for Implementation,” Diagnostics, vol. 8, no. 1, p. 20, 2018.
M. Muhammad, D. Zeebaree, A. M. A. Brifcani, J. Saeed, and D. A. Zebari, “Region of Interest Segmentation Based on Clustering Techniques for Breast Cancer Ultrasound Images: A Review,” Journal of Applied Science and Technology Trends, vol. 1, no. 3, pp. 78–91, 2020.
Y. Wang, et al., “Deeply-Supervised Networks with Threshold Loss for Cancer Detection in Automated Breast Ultrasound,” IEEE Transactions on Medical Imaging, vol. 39, no. 4, pp. 866–876, 2020.
R. Guo, G. Lu, B. Qin, and B. Fei, “Ultrasound Imaging Technologies for Breast Cancer Detection and Management: A Review,” Ultrasound in Medicine & Biology, vol. 44, no. 1, pp. 37–70, 2018.
X. Zhang, et al., “A Multicenter Hospital-Based Diagnosis Study of Automated Breast Ultrasound System in Detecting Breast Cancer among Chinese Women,” Chinese Journal of Cancer Research, vol. 30, no. 2, pp. 231–239, 2018.
M. A. Mohammed, et al., “Neural Network and Multi-Fractal Dimension Features for Breast Cancer Classification from Ultrasound Images,” Computers & Electrical Engineering, vol. 70, pp. 871–882, 2018.
Breast Ultrasound Images Dataset, https://www.kaggle.com/datasets/aryashah2k/breast-ultrasound-images-dataset.
A. Raza, et al., “Deep Breast Cancer Net: A Novel Deep Learning Model for Breast Cancer Detection Using Ultrasound Images,” Applied Sciences, vol. 13, no. 4, pp. 2082–2082, 2023.
L. Shen, et al., “Deep Learning to Improve Breast Cancer Detection on Screening Mammography,” Scientific Reports, vol. 9, no. 1, 2019.
M. S. K. Inan, et al., “Deep Integrated Pipeline of Segmentation Guided Classification of Breast Cancer from Ultrasound Images,” Biomedical Signal Processing and Control, vol. 75, p. 103553, 2022.
M. Ragab, et al., “Ensemble Deep-Learning-Enabled Clinical Decision Support System for Breast Cancer Diagnosis and Classification on Ultrasound Images,” Biology, vol. 11, no. 3, p. 439, 2022.
G. Ayana, et al., “A Novel Multistage Transfer Learning for Ultrasound Breast Cancer Image Classification,” Diagnostics, vol. 12, no. 1, p. 135, 2022
K. Jabeen, et al., “Breast Cancer Classification from Ultrasound Images Using Probability-Based Optimal Deep Learning Feature Fusion,” Sensors, vol. 22, no. 3, p. 807, 2022.

No competing interests reported.

Breast Cancer Prediction Using Machine Learning: A YOLOv8 Approach

Status:

Version 1

Abstract

Figures

1. INTRODUCTION

2. RELATED WORK

3. METHODOLOGY

3.1. Data Collection and Preprocessing

3.2. YOLOv8 Model Architecture

3.3. Web-based Accessibility: Convenience at Your Fingertips

3.4. Training of the neural network model

3.5. Evaluation Metrics

4. RESULTS AND DISCUSSION

5. CONCLUSION

Declarations

Author Contribution

Acknowledgement

References

Additional Declarations

Status:

Version 1