University Hospital for Radiology – Research report

302010, 302013, 302043, 302044, 302048, 302053, 302055, 302070, 302071, 302074, 302075, 302079

artificial intelligence, dose management, DTI, dual-energy-CT, fMRI, High-field MRI, interventional neuroradiology, MR-spectroscopy, multimodal imaging, and quantitative MR-imaging

Ultrasound (elastography, musculoskelettal, peripheral nerves, ultrasound guided interventions)
Multiparametric imaging of prostatic cancer
MRI (quantification of fat/iron; cardiac MRI; spectrosopy of myocardium; musculo Skelettal MRI including stress MRI of the hip and knee, MR mammography)
Mikro CT
Cardiac CT
Dual Energy CT
Emergency Radiology, trauma, sports injury
Interventional Radiology (endovascular/oncology)
Real time dose monitoring of patients
Diagnosis and treatment of HCC (stereotactic RFA, TACE, SIRT)
Digital imaging/PACS/postprocessing of imaging data, artificial intelligence
Clinical trials

For clinical routine and research, the department is equipped with state-of-the-art imaging systems: 7 CT scanners (2 dual-source/64-slice/40-slice [sliding gantry]), 5 MRI scanners (2x 3T, 3x 1.5T), 2 PET-CT scanners (in cooperation with the Department of Nuclear Medicine), 3 angiography suites (1 biplane) and 15 high-end ultrasound systems. One sliding gantry CT operates in an imaging suite dedicated to stereotaxis and CT-guided procedures. The department has operated fully digitally since 1999, using a comprehensive imaging archive (PACS).

There are 69 staff members, including 31 residents (radiologists in training).

The Experimental Radiology section (https://radiologie.tirol-kliniken.at/page.cfm?vpath=gemeinsame-einrichtungen/exprad) is staffed by 8 physicists/mathematicians with different areas of interest: MRI, MRS, image data processing, radiation protection, deep learning and computer applications.

The department houses research facilities for animals (high-resolution RF coils for MRI, access to PET imaging). Large animals can be imaged on clinical CT and MR scanners. The core facility of Micro-CT (https://radiologie.tirol-kliniken.at/page.cfm?vpath=radiologie/forschung/micro-ct) is equipped with 2 micro-CT scanners, which are operated in cooperation with the Department of Orthopaedics and Traumatology. Both scanners can be used for high-resolution imaging of small animals and one can also be used for high-resolution scanning of extremities, to evaluate patient bone density.

Our research projects are mostly interdisciplinary and clinically oriented. We focus on rapid translation of research results into clinical practice. The two departments have a long-standing PhD programme (Image-Guided Diagnosis & Therapy) and a clinical PhD programme (Clinical Imaging Science). 2021 a FWF grant for the “Image Guided Diagnosis and Therapy” programme (DOC 110-B) was awarded. Within this programme together with the Departments of Nuclear Medicine, ENT, Radiation Therapy and Onkology and Medical Physics the research expertise was enlarged by mathematics, computer science and engineering including partners from the University of Innsbruck LFU (Applied Mathematics and Computer Science) and from the Tyrolean Private University UMIT (Biomedical Engineering).

CT imaging

Gerlig Widmann

a) Head and Neck imaging

Use of AI for classification of cervical lymph nodes in head and neck cancer
Frequency an consequences of cervical lymph node overstaging in head and neck carcinoma

b) Thoracic imaging

Quantification of lung injury 1 year after COVID-19 pneumonia.
Evaluation of AI-based software for lung nodule detection, coronary calcium evaluation, measurement of aortic diameters, emphysema quantification and pneumonia quantification
Deep learning-based texture analysis and quantification in interstitial lung disease

Fig. 1: Deep learning-based texture analysis of combined pulmonary emphysema and fibrosis. Images show overlay of the texture features on the computed tomography scan.

c) Abdominal imaging

Dual Energy CT based evaluation of diffuse liver disease and HCC: fibrosis and fat quantification
Dual Energy CT based quantification of iron content in liver
Structured reporting as a basis for tumor registry for radiomics research

d) Low dose imaging

Identification and characterization of patients being exposed to computed-tomography associated radiation-doses above 100 mSv.
Development of an ALADA reference quality approach in craniofacial trauma CT
Clinical verification of an ALADA protocol in craniofacial trauma
Prediction of ALADA Doses for Implant Site Analysis Using Slope of Hounsfield Units- external collaboration with King Saud University

Experimental Radiology

a) Image Processing

Wolfgang Recheis

Image processing and analysis, including rapid prototyping based on radiological data, represent core interests and tasks, including multidimensional visualisation, the quantification of disease patterns based on texture analysis, shape analysis and others. Moreover, our new micro-CT core facility allows depiction of structures on μm scale in all three spatial dimensions.

Fig. 2: Full-colour 3D printout of an arteriovenous malformation (AVM) on the basis of 2D angiography, mapped on 3D CT

Fig. 3: Screenshots of our in-house developed dose management system “SumDose”: a real-time dose-monitoring system, work-in-progress of the dose management group

Small Animal MRI

Christian Kremser

In addition to clinically oriented projects, the Department of Radiology conducts experimental MR examinations on various animal models and specimens. Over the years, coil setup and imaging sequences have been optimised accordingly. Study topics include: mouse spinal cord imaging after shock wave treatment to promote spinal cord repair; volumetry of mouse myocardium to quantify fibrotic changes after constriction of the aorta and pulmonary artery; volumetry of the mouse brain to quantify fixation shrinkage; pulse wave velocity imaging in rabbits to evaluate aortic plaques; volumetric quantification of subcutaneous and visceral fat in mice on different diets; detection of intervertebral disc herniation in knockout mice.

Fig. 4: Examples of mouse brain MR images obtained in 3T with an isotropic resolution of 200µm.

Fig. 5: Examples of mouse whole-body MR images obtained in 3T for quantification of subcutaneous and visceral fat.

MedCorpInn– Retrospective Intersectional Corpuslinguistic Analysis of Radiological and Medical Reports of Medical University of Innsbruck

Claudia Posch and Karoline Irschara (University of Innsbruck), Leonhard Gruber and Stephanie Mangesius (Medical University of Innsbruck)

This project is compiling a large linguistic corpus (digitalised data collection) of radiology reports and medical reports, in order to conduct corpus and discourse linguistic as well as gender medicine research. After standardisation, automated data processing and anonymisation, the radiology corpus contains over 5 million anonymised radiology reports, which can be queried using corpus linguistic and NLP tools. With respect to linguistic patterns and differences, this unique data set can be used e.g. to study whether differences in language use are connected with social and demographic factors.

Abdominal MR Imaging

Michaela Plaikner, Benjamin Henninger, Christian Kremser

Morphological and functional MRI in all-organ systems development of novel MRI applications and MR sequences. Examples of research projects: fat, iron or combined disease; influence of iron on the evaluation of liver fat (Fig. 5).

a) MRI for the evaluation of diffuse liver disease:

evaluation of different MRI methods (relaxometry, chemical shift imaging, multi-echo approach, Dixon screening), in order to detect diffuse liver disease (fat, iron or combined disease); influence of iron on the evaluation of liver fat (Fig.6).

Fig. 6: Left: 43year old patient with fatty liver.

b) MR-Elastography (MRE)

MRE is increasingly used in hepatic MRI to detect and classify fibrosis in the early stages before morphological changes have occurred. In our department, MRE is already integrated into the routine hepatic MRI protocol and used for various research projects.

Fig. 7: Example of hepatic MRE. The colour image represents the stiffness map obtained. In this case, the mean hepatic stiffness was 4.83kPa, which indicates grade F3 fibrosis.

Cardiovascular MR Imaging

Agnes Mayr, Christian Kremser, Paulina Poskaite, Mathias Pamminger, Felix Troger, Matthias Schwab in cooperation with the Department of Cardiology

a) STEMI CMR: CMR Parameters of Myocardial Tissue Damage in ST-Elevation Myocardial Infarction (STEMI).

Since 2005, almost 900 patients have been examined under a comprehensive cardiac MRI (CMR) protocol within the first week as well as 4 months, 12 months and 10 years after acute STEMI. In more than 60 internal original papers, CMR myocardial infarction severity markers were assessed and the effects of CMR on optimised risk assessment shortly after STEMI were evaluated.

b) TAVI CMR: CMR to Guide Transcatheter Aortic Valve Implantation (TAVI).

This ongoing randomised study investigates the non-inferiority of TAVI CMR to TAVI CT for the first time, with regard to efficacy and safety end-points in the guidance for TAVI evaluation.

c) 4D Phase Contrast Flow Measurements

Patients with different grades of aortic valve stenosis: comparison of 4D flow-assessed stenosis severity with 3D echocardiography and invasive measurements.
Patients with cryptogenic stroke: recording turbulent kinetic energy, changes in flow patterns and regional wall stresses as well as occurrence of a vortex-shaped flow along the thoracic aortic wall, in order to optimise the elucidation of potential cardioembolic sources.

Fig. 8: A proposed segmentation pipeline. Firstly a 2D U-Net is trained on the individual images of the training dataset. After that a 3D U-Net is trained to refine the segmentation and return the final three-dimensional segmentation volume for the left ventricle. Input for the 3D CNN are the LGE CMR volumes as well as the possibly pertubated segmentation masks for scar and MVO provided by the pre-trained 2D CNN. — Fig. 8: A proposed segmentation pipeline. Firstly a 2D U-Net is trained on the individual images of the training dataset. After that a 3D U-Net is trained to refine the segmentation and return the final three-dimensional segmentation volume for the left ventricle. Input for the 3D CNN are the LGE CMR
volumes as well as the possibly pertubated segmentation masks for scar and MVO provided by the pre-trained 2D CNN.

d) Deep learning based automatic scar segmentation on late enhancement contrast enhanced MR

This ongoing project aims to develop a fully automatic framework to process 3D cardiac MR datasets for quantification of infarct scars. First results have already achieved results which outperform published state-of-the-art methods.

Fig. 9: Example of fully automated scar detection

e) Error correcting 2D-3D cascaded network for myocardial scar segmentation on LGE CMR images

In this work, we develop a cascaded framework of two-dimensional and three-dimensional convolutional neural networks that should enable to calculate the extent of myocardial infarction in a fully automated way.

Musculoskeletal Imaging

Andrea S. Klauser et al., Multicentre collaboration for indication in MSK imaging and interventions

Sonography of carpal tunnel: definition of cut-off values and correlaztion with Sonoelastography (Fig. 12)
Sonoelastography of epicondylitis, plantar fasciitis and Achilles tendon: accuracy in comparison with histology
Sonographically guided (SG) injections in CTS: sonoelastographic appearance
SG peripheral nerve injection: sonomorphology
SG nerve entrapment syndromes in comparison with surgical outcome
SG injection in sacroiliac joints of children: to prove feasibility
Dual-energy CT (DECT) in gout: comparison with US- findings, Cardio-DECT in gout and subgroup patients (Fig.13)
X-ray in comparison with DECT in gout patients
X-ray in comparison with US in erosion assessment
MRI in comparison with US in tendon overuse assessment
MRI whole-spine imaging in inflammatory disease: therapeutic follow-up assessment
MRI sacroiliitis: therapeutic follow-up assessment
MR tractography (DTI, ADI) in median nerves of healthy volunteers and CTS patients: comparison with sonography.

Fig. 10: Left: Gout deposits (in green) detected by DECT; Right: Gout deposits (green) in coronary arteries detected by DECT

Non-Invasive Cardiac/Cardiovascular CT Imaging

Gudrun Feuchtner

a) Coronary heart disease:

atherosclerosis, high-risk plaque marker, risk estimation, algorithms, non-invasive FFR

b) Structural heart disease CT

for planning TAVR, TMVR and catheter ablation, including 3D printing, advanced 3D visualisation, congenital heart disease (Fig. 12).
CT: 10 Multicentre trials

Fig. 12: 3D Print of coronary arteries after image post-processing.

c) Atherosclerotic Burden and its Relevance to Different Diseases and Treatment Strategies.

Bernhard Glodny, Johannes Petersen

Cardiovascular and cerebrovascular sequelae of atherosclerotic disease are among the leading causes of morbidity and mortality in humans. Atherosclerosis can be detected using different modalities of diagnostic imaging. Moreover, treatments are planned by means of clinical imaging, and cardiovascular risk profiles can be compiled individually. CT can be used to quantify the “atherosclerotic burden” of all vascular territories in an objective and reproducible manner.

Relationships between oral health and health in general have been suspected for many years and could also be proven by imaging studies, with evidence of an association between inflammatory changes in the periodontium and increased arteriosclerosis load, thus supporting the concept of arteriosclerosis as an inflammatory disease.

Ultrasound

Hannes Gruber, Alexander Loizides

Peripheral nerve sonography
SG interventions in the peripheral nervous system
SG carpal tunnel release
Sonographic evaluation of soft tissue masses (incl. MSK contrast-enhanced sonography (CEUS))
Sonography of the musculoskeletal system
Sonographic assessment of functional musculoskeletal disorders
SG pain therapy
Angiologic assessment and SG interventions in the vascular system

The surgical ultrasound section is a leader in the development of ultrasound techniques for the evaluation of peripheral nerves, ultrasound-guided nerve root infiltration and pain therapy. One of the most recent publications illustrates our work.

The axillary nerve (AN) is frequently injured during shoulder trauma and imaging is required to define the site and extent of nerve injury. However, the AN has a rather complex course through several soft tissue compartments of the shoulder and axilla. Detection and sonographic assessment therefore require thorough knowledge of the local topography.

Fig. 13: The 2 US scans a (longitudinal and axial) show the sonographic situation at the “departure” of the AN from the plexus, the 2 US scans b (longitudinal and axial) show the sonographic situation at the “middle segment”, and the 2 US scans c (longitudinal and axial) show the sonographic situation at the “distal segment”. The white arrowheads indicate the AN, the white arrow indicates the posterior fascicle of the plexus, (*) indicates the axillary artery, (^) indicates the cortex of the humeral head, (Δ) indicates the subscapular muscle, the rings (0) indicate the teres minor muscle, (~) indicates the long head of the triceps muscle, and the 3 bowed white arrows indicate the 3 terminal branches of the AN forming within the axillary gap.

Our investigation aims to define reliable, anatomical landmarks for AN sonography in volunteers and later to validate the proposed sonographic examination protocol in patients. With strict adherence to the proposed examination algorithm, sonography of the AN was feasible in all volunteers and patients. The findings correlated nicely with the gold standard of “surgical exploration” in respect of the severity and topography of neural impairment.

Based on our study results, we propose our algorithm for AN sonography as the first-line imaging tool for the assessment of axillary nerve trauma (PIC).

Fig. 14: Carpal tunnel syndrome: diagnosis by means of median nerve elasticity—improved diagnostic accuracy of US with sonoelastography (Miyamoto et al., 2014).

Interventional Oncology/ Stereotaxy and Robotics

Reto Bale

Image-guided tumour ablation
Stereotaxis
Robotics

Targeting
Interventional oncology

Stereotactic ablation of liver tumours:

Radiofrequency ablation (RFA) allows local curative tumour treatment by inducing coagulation necrosis with a high-frequency alternating current. The major limiting factor of conventional US- and CT- guided percutaneous single probe ablation is the tumour size. Navigation systems allow for 3D-planning of multiple overlapping ablation zones on the CT datasets and precise transformation into the real patient. We developed the worldwide first aiming device for frameless stereotactic punctures and performed the first in man stereotactic radiofrequency ablation (SRFA) of a liver tumor in 2001.

Meanwhile > 1200 patients with > 4500 liver tumors, most of them being inoperable, have successfully been treated at our department. Meanwhile our team has also trained the SMZ Ost in Vienna and the Ordensspital Barmherzige Brüder in Linz. In March 2023 the first stereotactic thermal ablation outside Europe was performed by Prof. Odisio at the MD Anderson in Houston, Texas, USA under supervision of Reto Bale.

Fig. 15: SRFA of a liver tumor with verification of precise needle placement by image fusion (screenshot of 3d navigation system)

Two PhD students from the “Imaged Guided Diagnostics and Therapy – Academic Research Triangle” (IGDT- ART) – funded by the FWF” program are supervised by Matthias Harders from the LFU and Reto Bale:

Adela Moravova: (Semi-)Automatic trajectory planning for multi-probe stereotactic thermal ablation of liver tumors

Stefano Fogarollo: Fast, non-rigid image fusion using deep learning for treatment evaluation after thermal ablation

Genderimaging

Imaging and Targeted Biopsy of the Genitourinary (GU) Tract

Friedrich Aigner

Multiparametric ultrasound of the scrotum
Multiparametric ultrasound of the prostate
Multiparametric MRI of the prostate
Fusion imaging of the GU-tract
Fusion targeted biopsy of the prostate

The simultaneous application of structural and functional imaging techniques is described as multiparametric (MP) (Fig. 16). Studies have shown that the MP approach results in greater diagnostic accuracy (Fig. 17).

The use of fusion imaging in uroradiology improves ultrasound lesion-detection rates, shows more reliable size controls at different time points, is an alternative to in-bore biopsies (Fig. 18) and can be used for focal therapy.

Fig. 16: Multiparametric MRI (a – c) and multiparametric transrectal ultrasound (d – f) of the prostate in the same patient show a histologically confirmed prostate carcinoma (arrows) as an area of low signal intensity / low echogenicity in T2-weighted MRI (a) and B-mode scan (d), with increased cell density in diffusion imaging (b, ADC-MAP) and ultrasound elastography (e, colour-coded blue), and as an area of hyperperfusion in contrast-enhanced MRI (c) and Doppler ultrasound (f).

Fig. 17: Patient with suspected testicular cancer: the multiparametric ultrasound of the testicle shows (a, arrow) a hypoechoic suspected cancerous lesion in B-scan, which is (b, arrow) soft in elastography (colour coded in green and red) and (c, d) avascular in (c) Doppler and (d) contrast medium ultrasound. In the light of these mpUS features, the new working diagnosis was changed to “cancer-simulating ischaemic orchitis”.

Fig. 18: US/MRI fusion-targeted biopsy of the prostate: in the left image, T2-weighted MRI shows the prostate carcinoma as an area of low signal intensity that also infiltrates the urethra (arrow); the right image shows the real-time ultrasound in B-mode on the split ultrasound monitor as well as the correct needle position (arrow) whilst performing the ultrasound-guided biopsy of the suspicious MRI lesion. Histology revealed carcinoma of the prostate with a Gleason score of 8 (4 + 4).

Imaging and Targeted Biopsy of Breast Cancer

Birgit Amort, Leonhard Gruber, Silke Haushammer, Martin Daniaux

The research team focusing on the diagnosis of breast cancer has a vast wealth of experience with all imaging modalities currently used for breast evaluation, including invasive methods such as fine-needle aspirations and/or percutaneous biopsies using stereotaxis or ultrasound guidance. Our unit serves as the largest screening and assessment centre of the national breast-screening programme in Tyrol. Approximately 10.000 mammograms and breast ultrasound studies are performed each year.

The last years have seen the impact of several new research items on daily clinical breast diagnostics routine: Dual-energy (DE) contrast-enhanced mammography is one of the latest developments in breast care. Imaging with contrast agents in breast cancer has been described in MRI and CT. However, high costs, limited availability and high radiation doses have led to the development of contrast-enhanced spectral mammography (CESM), which offers comparable breast cancer detection rates to MRI, yet at a lower false-positive rate, underlining its utility for preoperative local staging. Furthermore, CESM can be used in patients with MRI contraindications.

Increasingly, fusion sonography has been successfully introduced into daily clinical routine and is now employed in cases with sonography-occult findings in MRI and has been repeatedly demonstrated to avoid high-cost and stressful MRI biopsies. Our team has developed special experience to perform such fusion studies even using prone-position breast MRI studies by a more localized landmark-based approach.

Fig. 19: woman with an invasive ductal carcinoma, showing inhomogeneous breast tissue at 9 o’clock, as demonstrated by conventional mammography (a, b). Low-energy CESM images (c) are equivalent to conventional mammography, whereas spectral imaging shows a contrast-enhancing mass.

RVS ( Real time virtual sonographie) with MRI and CT

Data Management RVS Bioposies

RVS Biopsy vs. MRI Biopsy similar malignacy rate with 20- 29% (Benchmark: 20-50%)
RVS Biopsy: faster, more economical, more comfortable for the patient

Advantages RVS I

RVS enables to find most lesions
Short examination time (~ 10 min.)
Supine position same as standard examination position in ultrasound
Synchronization of two modalities in real time
FOV in supine Pos. < in prone Pos.
Biopsy chest wall and axilla near possible

Advantages RVS II

Biopsy under real time conditions
Selection for MR-biopsies
Significant increase of the detection rate of IELs
Tumor expansion (especially non mass lesions after chemotherapy) more exactly then in B-mode

Real time Ultraschall in Fusionsmodus- IndikationenSecond look-US in incidental enhancing breast lesions (IELs)

Management von km enhancenden Herdläsionen/ non mass Läsionen
Ultraschall gezielte Stanzbiopsie nach RVS

CT Imaging

Luger AK, Sonnweber T, Gruber L, Schwabl C, Cima K, Tymoszuk P, Gerstner AK, Pizzini A, Sahanic S, Boehm A, Coen M, Strolz CJ, Wöll E, Weiss G, Kirchmair R, Feuchtner GM, Prosch H, Tancevski I, Löffler-Ragg J, Widmann G. Chest CT of Lung Injury 1 Year after COVID-19 Pneumonia: The CovILD Study. 2022 Aug;304(2):462-470.
Santer M, Kloppenburg M, Gottfried TM, Runge A, Schmutzhard J, Vorbach SM, Mangesius J, Riedl D, Mangesius S, Widmann G, Riechelmann H, Dejaco D, Freysinger W. Current Applications of Artificial Intelligence to Classify Cervical Lymph Nodes in Patients with Head and Neck Squamous Cell Carcinoma-A Systematic Review. Cancers (Basel). 2022 Nov 2;14(21):5397.
Widmann G, Beyer A, Jaschke W, Luger A, Zoller H, Tilg H, Schneeberger S, Wolf D, Gizewski ER, Eder R, Torbica P, Verius M. Identification and characterization of patients being exposed to computed-tomography associated radiation-doses above 100 mSv in a real-life setting. Eur J Radiol Open. 2022 Dec 21;10:100470. doi: 10.1016/j.ejro.2022.100470. eCollection 2023.
Widmann G, Schönthaler H, Tartarotti A, Degenhart G, Hörmann R, Feuchtner G, Jacobs R, Pauwels R. As low as diagnostically acceptable dose imaging in maxillofacial trauma: a reference quality approach. Dentomaxillofac Radiol. 2023 Feb;52(3):20220387.
Widmann G, Dangl M, Lutz E, Fleckenstein B, Offermanns V, Gassner EM, Puelacher W, Salbrechter L. Can ultra-low-dose computed tomography reliably diagnose and classify maxillofacial fractures in the clinical routine? Imaging Sci Dent. 2023 Mar;53(1):69-75.