According to the 2022 World Cancer Report published by IARC (International Agency for Research on Cancer), breast cancer accounts for 13.6 percent of all cancer cases (male and female) in India. In women, it accounts for 26 percent of all cancer cases. In the United States, breast cancer accounts for about 30% of all new cancer cases in women.
New research suggests that artificial intelligence (AI) could help fight this dangerous disease. Early and accurate diagnosis could be crucial for treating patients, and a newly developed AI system promises to do just that with a nearly perfect diagnosis.
A research paper titled “Ensemble deep learning-based image classification for breast cancer subtype and invasiveness diagnosis from whole slide image histopathology,” published last month in the journal Cancer, details an AI model that classifies and identifies different types of breast cancer present in a patient, as well as ruling out malignant (cancerous) ones in the first place by identifying benign tumors.
The study, conducted by researchers at Northeastern University, Boston, and the Maine Health Institute for Research, developed an AI model that analyzes high-resolution histopathological (tissue-level microscopic) whole slide images of breast tumor tissue.
The AI system, which outperforms earlier machine learning (ML) models in this field by combining predictions from other ML models, is able to identify and classify tumours as malignant (cancerous) or benign (non-cancerous) using historical data fed to the model during training.
It was trained on publicly available datasets called BracHis (Breast Cancer Histopathological Database) and BACH (Breast Cancer Histopathology Images). For BACH, microscopic breast tissue images were carefully labeled by medical experts, classifying images into four categories – normal, benign, in situ carcinoma and invasive carcinoma.
Exemplary microscopy images demonstrating the four classes in the BACH dataset (Image source: Cancer 2024, 16(12), 2222)
And for BrakhIS, which contains 9,109 microscopic images of breast tumor tissue, it was used to classify benign and malignant tumors into 4 subclasses – malignant tumors were classified into ductal carcinoma, lobular carcinoma, mucinous carcinoma, and papillary carcinoma, and benign tumors were classified into adenoma, fibroadenoma, phyllodes tumor, and tubular adenoma.
Representative microscopy images of malignant and benign breast tissues from the BRAHIS dataset (Image source: Cancer 2024, 16(12), 2222)
Overall, the accuracy of the Ensemble ML model is 99.84 percent. Such a performance metric during the research and development phase shows optimistic promise for the real-world application of the technique.
“AI in biopsy cannot miss tumors and will get tired after diagnosing 10 or 20 people,” Saeed Amal told Northeastern Global News. Amal is a professor of bioengineering at Northeastern University and is leading the Ensemble Model Project.
Apart from diagnosis, AI systems have also made progress in prognosis and prediction of breast cancer-related prognosis. For example, AI can now predict neoadjuvant chemotherapy (NAC) response of breast cancer using hematoxylin and eosin (common stain in tissue imaging) images of pre-chemotherapy needle biopsies. The AI system responsible for this has an accuracy of 95.15 percent and is detailed in a paper titled “Development of multiple AI pipelines predicting neoadjuvant chemotherapy response of breast cancer using H&E-stained tissue”, published in the Journal of Pathology in May 2023.
In addition, AI has also made significant progress in identifying lymph node metastasis (spread of cancer cells through lymphatic nodes) and evaluating hormonal status which is crucial for breast cancer treatment. These and many other advances made by AI interventions in the last few years in the fight against breast cancer are reported in a review paper published in Diagnostic Pathology in February.
