Now that we are aware of the severity of TNBC, the advances in computational pathology and the potential of deep learning, is time to explore new possibilities and horizons to help with the diagnosis and prognosis of this type of breast cancer.
After reading the title of my thesis one could feel overwhelmed due to its length and the technicality of its words. So, I was when I first read it! However, after five months of working on it, it became way more understandable and coherent. And that is the purpose of this blog entry, to make the world understand in what does my thesis consist of.
For understanding it, the first thing that we need to understand is the role of breast cancer in our society. Breast cancer or breast carcinoma is the most common type of cancer to be diagnosed for women. To the extent where approximately the 12% of women in USA will be diagnosed with breast cancer during their lifetime. Particularly, about 10-20% of breast cancers test negative for both estrogen and progesterone receptors and excess HER2 in the lab, which means they are triple-negative (TNBC). Therefore, since hormones aren’t fueling the cancer’s growth, TNBC is unlikely to respond to hormonal therapy medicines and also to medicines that target the HER2 protein.
Not only triple negative is particular due to its difficulty of treatment, but also TNBC is considered to be more aggressive and have a poorer prognosis than other types of breast cancer, mainly because there are fewer targeted medicines that treat triple-negative breast cancer. Studies have shown that triple-negative breast cancer is more likely to spread beyond the breast and more likely to recur (come back) after treatment. Also, TNBC is more likely to be diagnosed in people younger than age 50, while other types of breast cancer are more commonly diagnosed in people age 60 or older. Furthermore, TNBC is more likely to be diagnosed in Black women and Hispanic women. In other words, it a type of breast cancer which is way more aggressive than other, it has a higher chance of recurrence and it is harder to treat. Therefore, the introduction of deep learning techniques seems a great opportunity to find out more about it to improve the diagnosis and prognosis of this severe carcinoma.
Now that we know a bit know of this particular type of breast cancer and the importance of its diagnosis and prognosis, following the title, we should try to understand what are the significant features that can be extracted from whole-slide images (WSI) of TNBC. The pathologic assessment of breast cancer tissue consists on the evaluation of biopsies and surgical specimens by the pathologists. This consists in a surgical excision where a sample of the tumor is collected in histologic glass slides, and thanks to sophisticated scanners, they can be digitally scanned. The result is a scanned WSI of the biopsies samples from the breast tumor of the patient.
Thanks to the advances of computational pathology, which is the analysis of digitized pathology images with associated metadata, typically using artificial intelligence (AI) methods, it is possible to improve the efficiency and accuracy of the evaluation process of WSIs. In recent years, several approaches have been used to analyze WSIs. The most of the models include detection, localization and segmentation of objects (i.e. histologic features) in these images. The recognizable and relevant histopathologic features include nuclear features, cellular/stromal architecture, or texture of the tissues. More particularly for breast cancer, relevant features include tumor regions, tumor Infiltrating lymphocytes (TILs), mitoses (typical and atypical), adipocytes, necrosis or fibrotic focus.
Example of TNBC WSI annotated by ESR11 that will be used to train the deep learning models
Finally, let´s see what is the role of deep learning in the extraction of features for TNBC. During the last years, deep learning, a type of machine learning, has been widely used for automatic image analysis due to the availability of large training datasets and the advancement of graphics processing units (GPUs). The possibilities are very wide with deep learning, applications range from classifying breast tissue slides to invasive or benign cancer, automatic grading of a carcinoma, survival prediction or treatment candidate proposal for the patients.
Concluding, now that we are aware of the severity of TNBC, the advances in computational pathology and the potential of deep learning, is time to explore new possibilities and horizons to help with the diagnosis and prognosis of this type of breast cancer. Different directions and approaches will be studied and explored during these three years of PhD. Particularly, due to the big size of a single WSI and its processing difficulties for GPUs, multi-magnification approaches will be explored for WSI analysis. Also, the combination of WSI analysis with deep learning and the addition of clinical data of the patients for the training of the models also presents a great opportunity for improving the accuracy and relevance of survival prediction algorithms or treatment candidate proposal systems.
Claudio Fernandez (ESR6) working on developing Deep Learning algorithms for helping the TNBC diagnosis and prognosis.
Claudio Fernández – ESR6.