Quirón-GE-URJC
Advanced Image Quantification Techniques
Using Machine Learning Tools
Aim
Integration of image quantification tools in the clinical workflow:
1. Seamless management of images, quantification and reporting.
2. Integration of standard open-source tools (i.e. Freesurfer).
3. Design and integration of in-house development image preprocessing and Deep
Learning pipelines.
4. Automatic quantification and reporting of every acquisition of a clinical study, to obtain
hospital-specific normative values and pathology characterization and detection.
2
Workflow
I
n
f
o
r
m
á
t
i
c
a
f
í
s
i
c
a
PACS
DICOM
SERVER
3
SERVER
4
SERVER
● Docker DICOM server
● One instance for each algorithm
○ Freesurfer
○ LST
○ Prostate
○ …
● Connects to MRI and receives images
● Sends PDFs to PACS server
Orthanc
5
SERVER
● Get DICOMS from Orthanc DICOM Server
● Processing DICOMS
○ Freesurfer
○ LST
○ Prostate
○ …
● Generating PDF files embedded in DICOMS
● Send PDF to PACS
Python3
6
SERVER
● Status and errors process
● Save output algorithms data
○ Freesurfer
○ LST
○ Prostate
○ …
MySQL
7
Volumetry Report (Freesurfer)
9
Volumetry Report (Freesurfer)
10
Prostate Tumor: Images and detection
11
Prostate zones and tumor: Segmentation
12
●Dense-U-Net Architecture for Segmentation
Prostate zones and tumor: Segmentation
13
Prostate: T2-DWI Registration
14
ITK-SNAP in-built registration Python pipeline (BRAINSFit)
registration
Prostate Tumor: Radiomics and Feature
Selection
15
●PYRADIOMICS → Feature Extraction
●Statistics → Feature Selection
Prostate Tumor: Classification
16
●SVM → 2 classes (CS).
●SVM (OVO) → 5 classes
(PI-RADS 1-5).
●Other models: LR, RF, NN, ...
Prostate Report
17
Prostate Report
18
