Histological analysis of lung tumor
tissue features
Yi Yin1,2, Irène Vignon-Clementel1,2, Dirk Drasdo1,2, Oliver Sedlaczek3, Kai Breuhahn3,
Arne Warth3, Benedikt Müller3, Peter Schirmacher3, Johannes Lotz4, Janine Olesch4
1. INRIA Paris-Rocquencourt, Le Chesney, France
2. Sorbonne Universités, Paris 6, Laboratoire Jacques-Louis Lions, Paris, France
3. University Hospital Heidelberg, Germany
4. Mevis Lübeck, Germany
Project: LungSys II, funded by BMBF (Federal Ministry of Education and Research, Germany)
2015-06-23
Outline
• About the lung cancer
– Lung cancer
– Histological classification
• Lung histological features analysis
–
–
–
–
Material: NSCLC tumor, pathology processing
Cell detection
Correlation between cell density and DWI MRI
Vessel reconstruction
• Conclusion
Lung cancer
Mortality rate for lung cancer




New cases world-wide 2012 – 1.8 million [1]
Death rate world-wide 2012 – 1.59 million [1]
Europe - highest incidence rate with 53.5 death per 100,000 inhabitants [1]
Costs for lung cancer treatment – 18.8 billion € [2]
[1] World Cancer Report, Edited by Bernard Stewart and Christopher P. Wild, ISBN 978-92-832-0429-9
[2] FERNANDEZ, R.L., et al. Economic burden of cancer Gross the European Union: a population based cost analysis.
The Lancet Oncology. 2013, Vol. 14, Iss. 12, pp. 1164-1174. ISSN 1470-2045.
Lung cancer
 Lung cancers are classified based on
histological type[3].
 small-cell lung carcinoma (SCLC)
 non-small-cell lung carcinoma (NSCLC)
 Lung Carcinoid Tumor
Non Small Cell Lung Cancer
Adenocarcinoma
40%
Squamous Cell
Carcinoma
25-30%
Large-Cell
Carcinoma
10-15%
Lung cancer histology provides a basis for diagnosis and therapy
[3] Lu, C; Onn A, Vaporciyan AA et al. (2010). "78: Cancer of the Lung". Holland-Frei Cancer Medicine (8th ed.).
People's Medical Publishing House.
NSCLC tumor histological image
B Müller / K Breuhahn / A Warth / O Sedlaczek
Priori clinical and
radiological indications
Pathology processing
Cutting and staining
Scanning
Hamamatsu NanoZoomer
Lung cancer: staining strategy
Cancer cell
Inflamatory cells
fibroblast
stroma
Blood vessel
KL1 stain: cancer cell
CD146 stain: blood vessel
 anti-CD146/MCAM (polyclonal, Atlas Antibodies, Stockholm, Sweden),
 and an anti-pan cytokeratin antibody (clone KL1, Abcam, Cambridge, UK).
Nuclei detection: image processing
Sub-image
Detected nuclei
Color deconvolution
Original image
Hematoxylin channel
KL1 channel
Segmentation result
Initial nuclei seed points
Mask image
Initial seed points detection
Nuclei segmentation
and classification
Different tumor tissues
Squamous cell carcinoma
Adenocarcinoma
#/mm2
Cell density
Cancer cell
Non-cancer cell
Squamous cell carcinoma
4400
2700
1700
Adenocarcinoma
2900
1600
1300
Cell density quantitative analysis
6000
5000
4000
3000
2000
1000
0
1
10
19
28
37
46
55
64
73
82
91
100
109
118
127
136
145
Cell number / mm2
Cell density
Index of slices
Tumor block
4500
4000
3500
3000
2500
2000
1500
1000
500
0
num_t
num_nt
1
8
15
22
29
36
43
50
57
64
71
78
85
92
99
106
Serial slices of tumor block
• Magnification: 20X
• Resolution: 0.456µm
• Size: 2.1 mm*2.1mm
Cell number/𝐦𝐦𝟐
Cancer and non-cancer cells
Index of slice
necrosis
DWI MRI
tumor
Normal lung tissue
Tumor block
Central necrosis (acellular)
Tumor tissue surrounding
central necrosis
Normal lung tissue
Tumor invading in
pericardial fat
D - value
 Signal intensity on DWI MRI
reflects the water molecules
diffusion in tissue
 Diffusion coefficient (D-value)
is estimated by using software
MITK-diffusion
0.006
How to predict the
micro information?
Cell density?
0.004
0.002
0
1
3
5
7
9
11 13 15 17 19
Index of pixel
D-values of pixels along the tumor center line
Cell density v.s Diffusion
DWI MRI diffusion estimation: O Sedlaczek
cut
DWI - MRI
Tumor slice
Tumor blocks
Correlation between DWI and cell density
6000
Cell Number / mm 2
5500
5000
L2
4500
L4
4000
Negative correlation
L1
3500
3000
2500
L5
2000
1500
1.3
1.4
1.5
1.6
1.7
D - values (10
1.8
-3
1.9
2
mm / s)
2
2.1
2.2
Vessel detection
Registration: J.s Lotz / J.Olesch
Resolution: 1.821µm
Magnification : ~=5X
Size: 4.5 * 4.5 mm2
Original image
Big vessel diameter:
~=30pixel *1.821 ~= 55µm
Binary result:
2D vessel cross-sections
Conclusion
• Quantitative analysis of lung histological tumor tissue
features
– Cancer and non-cancer cells segmentation
– Blood vessel detection and segmentation
• Study of correlation between DWI MRI and histology
– Diffusion coefficient estimation
– Cell density calculation of tumor blocks
Thank you for your attention
Any Questions?