Automated breast tumor ultrasound image segmentation with hybrid UNet and classification using fine-tuned CNN model

Abstract

Introduction: Breast cancer stands as the second most deadly form of cancer among women worldwide. Early diagnosis and treatment can significantly mitigate mortality rates.

Purpose: The study aims to classify breast ultrasound images into benign and malignant tumors. This approach involves segmenting the breast's region of interest (ROI) employing an optimized UNet architecture and classifying the ROIs through an optimized shallow CNN model utilizing an ablation study.

Method: Several image processing techniques are utilized to improve image quality by removing text, artifacts, and speckle noise, and statistical analysis is done to check the enhanced image quality is satisfactory. With the processed dataset, the segmentation of breast tumor ROI is carried out, optimizing the UNet model through an ablation study where the architectural configuration and hyperparameters are altered. After obtaining the tumor ROIs from the fine-tuned UNet model (RKO-UNet), an optimized CNN model is employed to classify the tumor into benign and malignant classes. To enhance the CNN model's performance, an ablation study is conducted, coupled with the integration of an attention unit. The model's performance is further assessed by classifying breast cancer with mammogram images.

Result: The proposed classification model (RKONet-13) results in an accuracy of 98.41 %. The performance of the proposed model is further compared with five transfer learning models for both pre-segmented and post-segmented datasets. K-fold cross-validation is done to assess the proposed RKONet-13 model's performance stability. Furthermore, the performance of the proposed model is compared with previous literature, where the proposed model outperforms existing methods, demonstrating its effectiveness in breast cancer diagnosis. Lastly, the model demonstrates its robustness for breast cancer classification, delivering an exceptional performance of 96.21 % on a mammogram dataset.

Conclusion: The efficacy of this study relies on image pre-processing, segmentation with hybrid attention UNet, and classification with fine-tuned robust CNN model. This comprehensive approach aims to determine an effective technique for detecting breast cancer within ultrasound images.

Keywords: Channel self attention; Deep learning; Hybrid attention module; Image preprocessing; Image segmentation; Spatial self attention.

Fig. 1

The expected total number of new cancer cases by cancer type for the year 2020 is shown in this statistic. It is estimated that there are around 18.1 million new instances of cancer worldwide, affecting people of all sexes and ages. The most common kind of cancer globally, affecting 2.2 million new cases, was breast cancer.

Fig. 2

(A) images from the original dataset, and (B) the normal class has been removed manually. In this dataset, every image has a ground truth, and every ground truth has the same dimension, color grading, and data format as the original image. A sample of a mask image is also shown.

Fig. 3

The CBIS-DDSM dataset comprises mammograms categorized into four classes, each exhibiting a range of artifacts across all classes. (A) Bening, (B) Bening Masses, (C) Malignant, (D) Malignant Masses.

Fig. 4

Merged Ground Truth: (A) shows the original images with multiple tumors. Ground truth (a), (b), and (c) are the ground truths of the original image (A), combining (a), (b), (c) gives the merged ground truth image (B).

Fig. 5

The workflow of this study is as follows: Frame (1) depicts some of the original ultrasound images of BUIS dataset. (2) Using different image preprocessing steps, a preprocessed dataset is generated. (3) The preprocessed dataset is segmented using segmentation techniques, resulting in a segmented dataset, produced by the UNet model. (4) Based on an ablation study, a more robust and accurate model is developed, and (5) the final model's results are analyzed using performance metrics.

Fig. 6

Image pre-processing steps.

Fig. 7

Text removal process, step by step.

Fig. 8

Base UNet like architecture.

Fig. 9

Hybrid attention UNet (RKO-UNet).

Fig. 10

Class distribution of training, validation, and test sets after splitting the segmentation dataset.

Fig. 11

Base CNN model.

Fig. 12

RkoNet-13 model.

Fig. 13

Test accuracy of all the ablation studies of UNet.

Fig. 14

Test accuracy of all the ablation studies of CNN model.

Fig. 15

Similarity between actual and predicted location of tumor. (A) Original Images, (B) Actual tumor location, (C) Predicted tumor location. (D) Tumor.

Fig. 16

Confusion matrix of the proposed model.

Fig. 17

Loss curve and accuracy curve over 100 epochs. (a) Depicted the accuracy curve of the training and validation, (b) depicted the loss curve of the training and validation.

Fig. 18

K-fold cross validation.