Recent Innovative Machine Learning-Based Techniques for Breast Cancer Diagnosis and Treatment

Editorial

Over the past few decades, the advancement of computational tools has been significantly changing the perspective of research, from almost human-based to almost machine-based, especially in the medical field, including diagnosis processes as well as treatment plans for various diseases and disorders. This editorial will highlight some of the recent breast cancer-related original research articles published in Technology in Cancer Research & Treatment (TCRT) journal, by Sage. The editorial will first discuss innovative machine learning-based techniques for breast cancer diagnosis, followed by techniques for breast cancer treatment.

Ge, et al ¹ developed a noninvasive technique for histological grading of invasive breast cancer based on extracting the radiomics features from ultrasound images, which are obtained noninvasively, as opposed to the commonly used biopsy-based histological grading technique, which is invasive and could result in post-biopsy complications. Their technique that showed to be capable of distinguishing between histological low grade and high grade is based on the presence of a relation between the intensity values in the ultrasound images and the histological grade as discussed in a previous study by Au, et al. ² In their study, Ge, et al ¹ obtained data from 383 patients at two independent sites, which was used for training and validation. From the noninvasively ultrasound images of the patients, 788 radiomics features were extracted and were then dimensionally reduced to 7 radiomics features. These features were used to train seven machine learning classifiers, from which the logistic regression classifier performed the best and hence they attempted to integrate it with a clinical factor, which is the size of the tumor, resulting in a combined histological grading model for invasive breast cancer. Both models; with and without the clinical factor, showed close performance that is promising, however building them in the future with a larger dataset could reduce any bias resulting from using a relatively small dataset, as well as confirming whether it is necessary to integrate clinical factors in the model or not. Adopting the technique of Ge, et al ¹ could facilitate the diagnosis and histological grading of invasive breast cancer, which in turns will aid in determining the right treatment plan, that is significantly dependent on the tumor histological grade.

The genetic-based research by Yi, et al ³ aimed at determining the subtypes of breast cancer by studying the endoplasmic reticulum (ER). Stress that affects the ER leads to the accumulation of unfolded proteins within the ER. This condition, which is known as ER stress (ERS) could contribute to tumor formation including breast cancer. Analyzing samples from both patients with breast cancer and without revealed that 8 ERS-related genes affect the survival prognosis. Unsupervised clustering for these 8 genes from 1109 patients with breast cancer resulted into obtaining two subtypes of breast cancer that are different in survival prognosis. Furthermore, a logistic regression-based model was trained using these 8 genes from samples obtained from 724 patients with breast cancer and 192 without. The constructed model was later tested on 475 subjects from which 375 had breast cancer and was found to have promising performance. The study also investigated the relation between these 8 genes and the immune microenvironment, but they did not reach a solid result, so it could be worth investigating this relation in a future study. Zhou, et al ⁴ investigated classifying breast tumors into benign (non-cancerous) and malignant (cancerous), using the Diagnostic Wisconsin Breast Cancer Database by Wolberg, et al, ⁵ which has 32-feature data for 569 subjects, from which 357 are labeled benign and 212 are labeled malignant. During their preprocessing stage, missing features were handled and normality test was performed to select a suitable feature selection method. Spearman correlation analysis was used to select the most 10 significant features. These features were fed into 7 different well-known classifiers, from which they found that the AdaBoost-Logistic algorithm worked the best in terms of accuracy. The authors presented a simple study that could help beginners to the field of machine learning to get a good overview of what is needed to build a machine learning-based diagnostic system.

In the following section, we will give an overview on some of the techniques used with breast cancer treatment that were published recently in TCRT journal. Since whole breast radiotherapy can reduce deaths caused from breast cancer, Ding, et al ⁶ studied how to handle position error resulting during therapy for two different intensity-modulated radiotherapy (IMRT) based modalities: the tangential IMRT (T-IMRT) and multi-angle IMRT (M-IMRT). Their study focused on patients with patients with left-sided breast cancer and they used data from 10 patients diagnosed with it. Tang, et al ⁷ focused on a different type of radiotherapy-related errors as they studied the errors resulting from having air gaps with T-IMRT. Data from 55 patients were used and air gap simulations with different depth were performed. They found that multivariable generalized estimating equations regression modeling could be used to describe the effect of the air gaps. Because breast cancer is highly heterogeneous at the molecular level Xie, et al, ⁸ studied the calcium-sensing receptor expression for both metastatic and non-metastatic breast cancer. They found that it has higher values in metastatic cases compared to non-metastatic cases, which makes receptor therapy for breast cancer more promising. Ocak, et al ⁹ studied the effect of combing Taxifolin and Epirubicin to treat breast cancer compared to using each of them individually. Their in vivo and in vitro studies revealed that combing both Taxifolin and Epirubicin is more effective for breast cancer treatment. Tang, et al ¹⁰ investigated the effect of intermittent fasting on triple-negative breast cancer cell progression and they found that cancer progression is affected by the intermittent fasting, from which targeted metabolomics could play a significant role in breast cancer treatment. Since treating breast cancer with radiotherapy can develop skin problems, regardless how accurate adjusting the radiation dose, Winkfield, et al ¹¹ explored using a keratin-based topical cream to manage the skin problems accompanying the radiotherapy. Their study included 24 patients and they found it promising to use keratin-based topical cream to handle radiation dermatitis. It is clear that a big portion of the current research attempts at handling problems associated with radiotherapy, which can eventually result in a significant improvement to the radiotherapy-based treatment.

In summary, the nine breast cancer-related research articles, published Technology in TCRT journal, by Sage that we discussed in this editorial tackled some of the common challenges that exist in breast cancer diagnosis and treatment, making it promising to have a better life for patients with breast cancer in the near future.