LE 6 Artificial intelligence for precision oncology in digital pathology – pathology
Artificial intelligence training robots (with examples) instead of programming to solve complex
problems
Digital pathology
- Suspect of breast cancer, take biopt, cut it in lab, stain and go to pathology
- Immunoscanners
Prognostic relevance of grading in cancer treatment
- Grade III, high grade, more aggressive, treat earlier
- Getting grade right is important for treatment
Neoadjuvant chemotherapy in TNBC (breast cancer)
- Diagnosis, then therapy, tumor should shrink, then surgery
- Half of patients do get benefits of therapy, half of patients have same size of tumor when
going into surgery
AI for image classification/digital pathology
- Set of training data (tumor against stroma), program takes multiple inputs and through
interconnections it produces one output
Pipeline in digital pathology department
- Scanning
o Artifacts pen marker, tissue fold, dust, ink
o With AI, is quality good?, if percentage of artifacts is small, then send to pathologist
- Assign slide to pathologist
o Form population screening of colon polyps and biopsies, about 70% is non-
informative, sort them and pathologist can focus on the 30%
o Train AI that can generate parts of report so pathologist does not have to do this
- Diagnosis
o Slide goes to pathologist
o Mitotic count, look for proliferated tumor cells in entire slide
o Pre read slide, and compute mitosis density, so pathologist can look at hot spot
- Treatment
o Mitotic density identified by AI can be used as biomarker
o Predicting responders and non-responders
o Making attention map with AI
Pathology – an introduction
Pathology = the study of the causes and effects of disease or injury
- Specific: the branch of medicine that deals with the laboratory examination of sample of body
tissue for diagnostic purposes
Breast cancer presence of calcification is a suspect, biopt is necessary
Lymphoma / Metastases / Infection Large lymph nodes, distinguish under microscope
Cancer treatment:
- Surgery
- Radiation therapy
- Chemo therapy
- Immuno therapy
, - Hormone therapy
- Targeted therapy
- Stem cell therapy
Recognition of tissue types/metaplasia:
- If epithelial cells are placed at tissue where it should not be, it could be cancer
- Sweat glands can look like lymph nodes when there is an inflammation
Cancer in the picture: pathology
Cell with mutation
Hyperplasia = increase in cells
- More cells means needing more place
- In bowel, there is fixed space, crypts are star
shaped because there is not enough space
Dysplasia = accumulation of mutations
- Disturbed maturation
- Pseudo stratification
- Loss of goblet cells
- Loss of nuclear orientation
- Nuclear atypia nuclei are bigger and not one shape, pink dots are nucleoli sign of active cell
division, mitotic figures
- Formation of nucleoli
- Abnormal crypt formation
In situ cancer
Invasive cancer = does not stay at same location, invasive growth
Next-generation DNA sequencing
1. Library preparation
2. Clonal amplification
3. Cyclic array sequencing
BRAF mutation targeted therapy
- Present in 10% of bowel cancer
- Is already present in premalignant diseases, hyperplasia
Prognosis prediction: biomarkers
- Lymph node metastases, if lymph node has epithelial cells it looks like cancer
- Extramural Vascular invasion, in thick blood vessel there is a tumor, tumors in vessel bowel wall
increased metastasis
Artificial intelligence training robots (with examples) instead of programming to solve complex
problems
Digital pathology
- Suspect of breast cancer, take biopt, cut it in lab, stain and go to pathology
- Immunoscanners
Prognostic relevance of grading in cancer treatment
- Grade III, high grade, more aggressive, treat earlier
- Getting grade right is important for treatment
Neoadjuvant chemotherapy in TNBC (breast cancer)
- Diagnosis, then therapy, tumor should shrink, then surgery
- Half of patients do get benefits of therapy, half of patients have same size of tumor when
going into surgery
AI for image classification/digital pathology
- Set of training data (tumor against stroma), program takes multiple inputs and through
interconnections it produces one output
Pipeline in digital pathology department
- Scanning
o Artifacts pen marker, tissue fold, dust, ink
o With AI, is quality good?, if percentage of artifacts is small, then send to pathologist
- Assign slide to pathologist
o Form population screening of colon polyps and biopsies, about 70% is non-
informative, sort them and pathologist can focus on the 30%
o Train AI that can generate parts of report so pathologist does not have to do this
- Diagnosis
o Slide goes to pathologist
o Mitotic count, look for proliferated tumor cells in entire slide
o Pre read slide, and compute mitosis density, so pathologist can look at hot spot
- Treatment
o Mitotic density identified by AI can be used as biomarker
o Predicting responders and non-responders
o Making attention map with AI
Pathology – an introduction
Pathology = the study of the causes and effects of disease or injury
- Specific: the branch of medicine that deals with the laboratory examination of sample of body
tissue for diagnostic purposes
Breast cancer presence of calcification is a suspect, biopt is necessary
Lymphoma / Metastases / Infection Large lymph nodes, distinguish under microscope
Cancer treatment:
- Surgery
- Radiation therapy
- Chemo therapy
- Immuno therapy
, - Hormone therapy
- Targeted therapy
- Stem cell therapy
Recognition of tissue types/metaplasia:
- If epithelial cells are placed at tissue where it should not be, it could be cancer
- Sweat glands can look like lymph nodes when there is an inflammation
Cancer in the picture: pathology
Cell with mutation
Hyperplasia = increase in cells
- More cells means needing more place
- In bowel, there is fixed space, crypts are star
shaped because there is not enough space
Dysplasia = accumulation of mutations
- Disturbed maturation
- Pseudo stratification
- Loss of goblet cells
- Loss of nuclear orientation
- Nuclear atypia nuclei are bigger and not one shape, pink dots are nucleoli sign of active cell
division, mitotic figures
- Formation of nucleoli
- Abnormal crypt formation
In situ cancer
Invasive cancer = does not stay at same location, invasive growth
Next-generation DNA sequencing
1. Library preparation
2. Clonal amplification
3. Cyclic array sequencing
BRAF mutation targeted therapy
- Present in 10% of bowel cancer
- Is already present in premalignant diseases, hyperplasia
Prognosis prediction: biomarkers
- Lymph node metastases, if lymph node has epithelial cells it looks like cancer
- Extramural Vascular invasion, in thick blood vessel there is a tumor, tumors in vessel bowel wall
increased metastasis
