H&E (Haemotoxylin and Eosin) : H&E staining is the most commonly used technique to visualize the tissue biopsy sections. H&E technique uses two different types of dye; one stains to the DNA in the nucleus of the cell (purple color) whereas the other stains the cytoplasm (parts of cell outside the nucleus) (pink color). H&E stained tissue slides are generally used for quality control purposes so as to assure that the tissue biopsy and FFPE preparation is correctly carried out.
H&E staining technique properly helps visualize (under microscope) different cells and structures within a tissue. The pathologist studies these slides to assess the following
- If the biopsy tissue sample is benign (non-cancerous) or malignant (cancerous)
- If the biopsy is malignant, the pathologist also assigns the tumor grade (i.e. pathological staging) based on the histological examination (Grade 1, 2, 3, or 4; in the increasing order of aggressiveness). These tumor grade could potentially serve as one of the clinical-data features in the machine learning algorithm for tissue classification. Note that pathological staging is different that clinical staging.
- The pathologist also determines if the tumor has likely metastasized (i.e. spread to other parts of the body). The pathologist may ask for more biopsy test if need.
To do:
- [types of biopsy (effects the quality of histopathology images)]
- [what does a pathologist do with the biopsy samples and examining histopathology images]
- [basics of breast cancer histopathology. This will form a solid foundations for the material present below]
<under_progress: raunak>
Histopathology
Important points in histopathology
Diagnostic and Understanding Challenges and opportunities
- Identifying tumors in lymph nodes in Hematoxylin-eosin (H&E) stained histopathological images is generally laborious and error-prone, esp. for small tumor foci. Immunohistochemical (IHC) staining can improve sensitivity but increases workload, costs and delays reporting. AI can achieve comparable performance to pathologists in H&E stained images and help them improve morphologic detection of metastatic tumors in lymph nodes \cite{Liu2019}.
- Mitosis count for prognosis: write about it
- Predicting invasive vs. non-invasive (tumor vs. not tumor) is relatively less challenging compared to distinguishing the sub-types in non-invasive ones: benign, atypia, Ductal Carcinoma In Situ (DCIS) where the diagnostic disagreements are remarkably high among pathologist \cite{Mercan2019}. The authors therefore propose a multi-stage approach (semantic segmentation into 8 tissue labels followed by feature extraction and diagnostic classification) for classifying the breast lesions into four classes: benign, atypia, DCIS, and invasive breast cancer. They design hierarchical scheme with a single diagnosis at a time as follows: (1) invasive vs. noninvasive diagnosis, (2) if noninvasive, classify into benign vs pre-invasive lesions, and (3) if pre-invasive, classify into DCIS or atypia. The main limitation here is that the Regions of Interests (ROI) have to be manually selected from Whole Slide Images (WSIs) before feeding each ROI into the multi-stage classifier pipeline.
- In \cite{Gecer2018}, the authors provide a method to first automatically identify ROI, then classify into five diagnostic categories of ductal proliferations: non-proliferative changes, proliferative changes, atypical ductal hyperplasia (ADH), ductal carcinoma in situ(DCIS), and invasive ductal carcinoma (IDC). Question: Are non-proliferative vs proliferative eqv. to benign vs. malignant or invasive vs. non-invasive? The authors record the zoom activity of pathologists to create ground truth data for ROIs.
- In \cite{Couture2018}, the authors propose automatically analysing H&E stained breast tumor microarray (TMA) images to identify patients that would benefit from molecular testing by predicting tumor grade, histologic subtype, estrogen receptor (ER) status, PAM50 intrinsic breast cancer subtype, and risk of recurrence score (ROR-PT). Although RNA-based, multi-gene molecular assays are available and widely used for patients with ER-positive/HER2-negative breast cancers, these tests are costly and unavailable in many countries. Thus, this kind of methods for inferring molecular subtype just from H&E stained images may reduce the number of patients sent for further testing suspecting benefit from further genomic tests, saving cost and time.
AI challenges and Potential Solutions
H&E slides variation
Tissue preparation, staining protocols and oxidation in the lab makes the whole slide images different when taken from different centers. Digitization by a camera introduces further variation. Transforming the color space and trying to get the color statistics of individual images towards a reference statistics can be helpful as suggested in page 861 of \cite{Liu2019}.
Grading
HE slides on lymph nodes biopsy
1. macrometastasis 2. micrometastatsis 3. isolated tumor cells (ITC) 4. negative Also useful to localize in the images micrometastasis or ITC cases.
Glossary
Lymph node: Also known as lymph gland, are ovoid or kidney-shaped organ of the lymphatic system which comprises of a lymphatic vessels carrying a clear fluid lymph. Lymph nodes are widely present throughout the body and are linked by lymphatic vessels. Lymph nodes condition play an important role in staging cancer which consequently decides the treatment plan and prognosis.
FNA: One of the three common biopsy techniques: fine-needle aspiration (FNA), vacuum-assisted biopsy and surgical biopsy.