From Microscope to Algorithm:
The Impact of AI on
Histopathology Practice
Dr. Muhammad Usman Shams
MBBS, M.Phil, FCPS (Histopathology)
Diploma in Healthcare Management
Objectives
From Microscope to Algorithms
The Journey
The Future
Utility of AI in Histopathology
The Opportunity
How AI Algorithm Works?
The Logic B
D
C
A
A From Microscope to Algorithm
1600-1800
1830-1990 2000...
1990…
Microscope
Photomicrograph
Whole Slide
Imaging
AI
A 400-year Journey
Timeline of Microscopy
Rome was not built in one day.
First MICROSCOPE First PHOTOMICROGRAPH (Not really)
The
Start
Wet Plate Collodian
Evolution of
Photomicrographs
(1845-1990)
Daguerreotype
Polaroid Film
Digital
Whole Slide Imaging
BLISS System (1994)
AI
Algorithms
The
Digital Era
How AI Algorithm works?
B
Soliman et al. Diagnostic Pathology (2024) 19:38
Computational Pathology
Five Different Types of Annotations
Computational pathology: A survey review and the way forward. Journal of Pathology Informatics (2024)
Algorithm
Convolutional Neural
Networks (CNN)
Machine Learning (ML)
Content based image
retrieval (CBIR)
Deep Learning (DL)
Artificial Intelligence
(AI)
AI in Histopathology
Let us experience AI
Reported by a Consultant Histopathologist
Reported by AI Algorithm
C Utility of AI in Histopathology
Computational pathology: A survey review and the way forward. Journal of Pathology Informatics (2024)
How AI is supporting Histopathologists?
“AI is especially helpful in the areas where pathologists do a lot of manual tasks, especially where we count cells,
we count nuclei, we count mitosis, we look for mitosis, we look for microorganisms such as H. pylori.”
Dr. Anil Parwani
Quantification
• Automated calculation of
breast biomarkers: ER, PR,
HER2
• Ki67 counting
• Mitosis detection
• Counting glomeruli in renal
biopsies
• Quantification of fibrosis &
fat
Diagnostic Assistance
• Finding DCIS in breast
• Gleason scoring of prostate
cancer
• Finding microorganisms
Daily Tasks
• Automatic quality check of
slides & staining
• Educational tool
• Image Matching (CBIR)
Novel Tumour Classification Schemes
• Prognostication based on epithelium to stroma ratio
• Risk stratification based on variations in TILs
• Correlation with other data (clinical, pathologic or genomic)
400+ CPath Diagnostic Tasks from 2018 to 2022
Journal of Pathology Informatics 15 (2024) 100357
Roche
• HER2 (4B5) - Breast
• HER2 FISH - Breast
• PDL1 (SP263) - NSCLC
Ibex
Galen™
• Prostate
• Breast
Path AI • AIM-PD-L1
IVD-approved AI Algorithms
Risk stratification of Colon Cancer based on variations
in TILs
Digital Pathology for Better Clinical Practice. Cancers (2024
How AI is challenging Histopathologists?
• ROI
• Reimburse-
ment
Cost
• Utility in real
world
• Cost :
Benefit ratio
Acceptance
• Data &
Storage
• Overfitting
• ‘Black box’
problem
Technology
• Quality of
Data
• Quality of
Slides &
Staining
Pre-Analytical
Variables
D The Future
Those who understand AI limitations and its uses, and overcome the
barriers to implementation, will be well poised to deliver the best
possible patient care now and in the future.
Thank you
It has been a privilege and honour.

The Impact of AI on Histopathology Practice

  • 1.
    From Microscope toAlgorithm: The Impact of AI on Histopathology Practice Dr. Muhammad Usman Shams MBBS, M.Phil, FCPS (Histopathology) Diploma in Healthcare Management
  • 2.
    Objectives From Microscope toAlgorithms The Journey The Future Utility of AI in Histopathology The Opportunity How AI Algorithm Works? The Logic B D C A
  • 3.
    A From Microscopeto Algorithm 1600-1800 1830-1990 2000... 1990… Microscope Photomicrograph Whole Slide Imaging AI A 400-year Journey
  • 4.
    Timeline of Microscopy Romewas not built in one day.
  • 5.
    First MICROSCOPE FirstPHOTOMICROGRAPH (Not really) The Start
  • 6.
    Wet Plate Collodian Evolutionof Photomicrographs (1845-1990) Daguerreotype Polaroid Film Digital
  • 7.
    Whole Slide Imaging BLISSSystem (1994) AI Algorithms The Digital Era
  • 8.
    How AI Algorithmworks? B Soliman et al. Diagnostic Pathology (2024) 19:38
  • 9.
    Computational Pathology Five DifferentTypes of Annotations Computational pathology: A survey review and the way forward. Journal of Pathology Informatics (2024)
  • 10.
    Algorithm Convolutional Neural Networks (CNN) MachineLearning (ML) Content based image retrieval (CBIR) Deep Learning (DL) Artificial Intelligence (AI) AI in Histopathology
  • 11.
    Let us experienceAI Reported by a Consultant Histopathologist
  • 12.
    Reported by AIAlgorithm
  • 13.
    C Utility ofAI in Histopathology Computational pathology: A survey review and the way forward. Journal of Pathology Informatics (2024)
  • 14.
    How AI issupporting Histopathologists? “AI is especially helpful in the areas where pathologists do a lot of manual tasks, especially where we count cells, we count nuclei, we count mitosis, we look for mitosis, we look for microorganisms such as H. pylori.” Dr. Anil Parwani Quantification • Automated calculation of breast biomarkers: ER, PR, HER2 • Ki67 counting • Mitosis detection • Counting glomeruli in renal biopsies • Quantification of fibrosis & fat Diagnostic Assistance • Finding DCIS in breast • Gleason scoring of prostate cancer • Finding microorganisms Daily Tasks • Automatic quality check of slides & staining • Educational tool • Image Matching (CBIR) Novel Tumour Classification Schemes • Prognostication based on epithelium to stroma ratio • Risk stratification based on variations in TILs • Correlation with other data (clinical, pathologic or genomic)
  • 15.
    400+ CPath DiagnosticTasks from 2018 to 2022 Journal of Pathology Informatics 15 (2024) 100357
  • 16.
    Roche • HER2 (4B5)- Breast • HER2 FISH - Breast • PDL1 (SP263) - NSCLC Ibex Galen™ • Prostate • Breast Path AI • AIM-PD-L1 IVD-approved AI Algorithms
  • 17.
    Risk stratification ofColon Cancer based on variations in TILs Digital Pathology for Better Clinical Practice. Cancers (2024
  • 18.
    How AI ischallenging Histopathologists? • ROI • Reimburse- ment Cost • Utility in real world • Cost : Benefit ratio Acceptance • Data & Storage • Overfitting • ‘Black box’ problem Technology • Quality of Data • Quality of Slides & Staining Pre-Analytical Variables
  • 19.
  • 20.
    Those who understandAI limitations and its uses, and overcome the barriers to implementation, will be well poised to deliver the best possible patient care now and in the future.
  • 21.
    Thank you It hasbeen a privilege and honour.

Editor's Notes

  • #6 First ever photomicrographs, 1845. This image was published by Alfred Donne and Leon Foucault in 1845 in French the medical textbook Cours de microscopie. Donne and Foucault took Daguerreotype photographs of specimens through a light microscope. The four images are: (fig 45) urea nitrate crystals, (fig 46) crystals of typhoid urine, (fig 47) crystals of uric acid, (fig 48) group of uric acid crystals.
  • #7 BLISS System (1994). The first digital microscope systems cost about $300,000 to set up and took over 24 h to scan a single slide. He worked with the American Board of Pathology to introduce virtual microscopy into board certification of pathologists. Company acquired by Olympus
  • #9 Five Different Types of Annotations. The most beneficial support of AI to pathology will be building up computational pathology on traditional histopathology. FOV: Field of View Computational pathology is a brand-new discipline that aims to enhance patient care by utilizing advances in artificial intelligence and data generated from anatomic and clinical pathology.
  • #10 AI: Intelligence exhibited by machines, particularly computer systems ML: Enables systems to learn from data without explicit programming DL: Uses multiple layers to learn complex understandings, inspired by neural network in humans CNN: Designed for image and video analysis, Spatial information & Extraction filters
  • #15 Distribution of diagnostic tasks in CPath for different organs from Table 9.11. This distribution includes more than 400 cited works from 2018 to 2022 inclusive. The x-axis covers different organs, the y-axis displays different diagnostic tasks, and the height of the bars along the vertical axis measures the number of works that have examined the specific task and organ
  • #17 Figure 2. Illustration of the DP-Immunoscore calculation method. (A) Densities of CD3+ and CD8+ at both CT and IM are converted into percentile values. The mean percentile of the four markers is calculated and represented into a five-category (IS0, IS1, IS2, IS3, IS4) or a three- (IS-Low, IS-Int, IS-High) or a two-category scoring system (IS-Low, IS-High). Based on measuring immune response at the tumor site, IS predicts the risk of relapse in localized colon cancer to identify patients who could be spared from chemotherapy. (B) Ring charts illustrating the relative contribution of each risk parameter to recurrence risk in patients with stages II and II/III colon cancer. IS (red) is the highest predictor of time to recurrence (TTR) in both subgroups.
  • #18 Overfitting” is when AI algorithms, trained on one dataset, have limited applicability to other datasets. The ‘black box’ problem is the inability of deep learning algorithms to demonstrate how they arrive at their conclusions.
  • #19 A STED (stimulated emission/depletion) micrograph image revealing actin (magenta) and microtubules (cyan) MUSE image of sebaceous gland A STED (stimulated emission/depletion) micrograph image revealing actin (magenta) and microtubules (cyan) of a young dissociated hippocampal neuron. Image by K. Jansen and E. Katrukha, Kapitein Lab, Molecular and Cellular Biophysics, Utrecht University, The Netherlands Where the light is focused from all sides to a common focus that is used to scan the object by 'point-by-point' excitation combined with 'point-by-point' detection Photon-tunneling microscopy[4] as well as those that use the Pendry Superlens and near field scanning optical microscopy “More than 80 percent of the time, patient biopsies will come back normal,” says Richard Levenson, professor and vice chair for strategic technologies in the UC Davis Department of Pathology and Laboratory Medicine. “Meanwhile, the patient is waiting for results for a potentially lethal disease. They live with a great deal of anxiety.” To remedy this, researchers at UC Davis are working on a new way to view tissue. They’ve created the MUSE microscope, which uses ultraviolet rather than visible light. MUSE samples do not require the rigorous preparation that can slow down analysis of conventional slides.