UK Imaging Informatics Group
Computer Aided Detection-CAD and VNA/... PreviousNext
UK Imaging Informatics Group > Questions & Answers > PACS Procurement & Business Cases >
Message/Author
 Link to this message Neelam Dugar  posted on Monday, October 31, 2016 - 03:21 pm Edit Post Delete Post Print Post
http://www.auntminnieeurope.com/index.aspx?sec=sup&sub=pac&pag=dis&ItemID=613546

I do think Computer Aided Detection(incorrectly called Computer Aided Diagnosis) will be a standard component of PACS in 2020. This combined with PACS acting as an XDS consumer accessing VNA data will transform the diagnostic capabilities of PACS.

My view is that CAD (detection) will aid radiologists in detection of abnormalities, and help radiologists make more accurate diagnosis. It will reduce number of missed shadows on CXR, missed lung nodules on CT, missed fractures, missed incidental PE etc.

The role of radiologists will move from "detection" to "interpretation". XDS consumer function of VNA will allow timely access to other types of medical data--blood results, clinic letters, discharge summaries, endoscopy reports, etc is key to radiologists to better interpretation of radiology images.
 Link to this message Neelam Dugar  posted on Tuesday, November 01, 2016 - 06:56 pm Edit Post Delete Post Print Post
https://www.sectra.com/medical/breast_imaging/solutions/ris-pacs/add-ons/3rd_par ty_providers/icad/index.html

http://www.healthimaging.com/topics/diagnostic-imaging/computer-aided-detection- improves-lung-nodule-detection

http://flex.flinders.edu.au/file/750e1e60-2401-42e8-8445-ba306204ed9a/1/Thesis-D onnelley-2008-04chapter3.pdf

http://www.medscape.com/viewarticle/568260_3

Some very interesting articles on Computer Aided Detection. Sectra has integrated Mammography CAD to their PACS.

Key aspect of integration will be
1. How efficient is the integration with the PACS reading workflow?
2. I think most radiologists would like to use it as a 2nd reader--when required.
a. If they have already seen the numerous lung mets--they wont want to apply the CAD.
b. If they don't see any lung nodules--they would want a CAD super-imposition with 1 click--AFTER the preliminary non-CAD display.
3. How many false positives does the CAD produce? As false positives will add to reading time and make us slower.

Where do I think CAD is likely to be useful:
1. Mammography--already happening
2. Chest Shadow/nodule--on Chest Xray
3. Lung nodules--on CT
4. Fracture detection

It will be a learning curve for radiologists knowing when they can trust CAD and when they cannot. They will need to understand the false negative and false positive rates of CADs. Interesting time ahead!

I think CAD is going to be the next big aspect to optimising radiologists reading workflow.
 Link to this message Neelam Dugar  posted on Sunday, November 20, 2016 - 01:48 pm Edit Post Delete Post Print Post
http://www.auntminnie.com/index.aspx?sec=sup&sub=adv&pag=dis&ItemID=100084

Another article on CAD (Computer Aided Detection) integration with PACS using Post-processing workflow oh IHE.

So from a workflow perspective-
1. Images will be sent from modalities to PACS
2. CAD software (integrated with PACS) will create a separate series with the CAD markings on them.
3. Radiologists will have access to CAD markings (INSTANTLY) if they wish to review them.

I think radiologists would benefit from CAD for
1. Mammography
2. CXR--lung shadows
3. Plain xray--fracture
4. CT Chest--lung nodules

I have used CAD in my clinical practice--CT lung nodule detection. Whilst I found it useful, I soon gave up, as it was too time consuming. I had to launch another application and then wait for CAD to do its calculation. Making CAD instantly available as part of reporting workflow is key to its uptake by radiologists.
 Link to this message Neelam Dugar  posted on Thursday, November 24, 2016 - 11:22 pm Edit Post Delete Post Print Post
Interesting Ben! I would be keen to what you are doing--if u plan to visit the U.K.

Workflow is key. This is how I envisage it happening.
When reporting a CXR my current image for reporting appears on my middle monitor and prior on my right monitor.
With CAD I would expect the "same" workflow but the additional CXR images with CAD markings would appear as a separate series within the same study.
This kind of workflow would make it usable--anything more time consuming than this is unusable.
However, DICOM standard will need to be an integral part of seamless CAD workflow.

CXR shadows
Mammography
Plain X-ray fracture detection
CT lung nodule detection

I can see CAD actually benefitting Radiologists and becoming an invaluable tool for reporting. Technology and Standards currently are lagging behind!
 Link to this message David Clunie  posted on Friday, November 25, 2016 - 04:04 pm Edit Post Delete Post Print Post
Generic post-acquisition processing workflow can be challenging, and depends a lot on whether the work is:

- pre-processed (i.e., done before the radiologist needs it)

- on demand (i.e., requested while the radiologist is reporting, +/- their interactive input)

Even in the pre-processed case, which sounds like a no-brainer, there are issues such as determining when a "study" has ended (if multiple images need to be pre-processed together).

The pragmatic (state of the art) Mammo CAD workflow is to process single images independently as they are pushed from the modality to the CAD server and thence the results (Mammo CAD SR) sent to the PACS. Even that simple approach presents problems (e.g., when is the CAD ready to trigger "ready to read" status for radiologist reading worklist).

The "managed" approach to pre-processing workflow (IHE Post-Processing Workflow (PWF) "http://wiki.ihe.net/index.php/Post-Processing_Workflow"), which uses a UPS worklist) has largely been ignored by the industry as adding complexity with too little (if any) benefit.

For on demand processing, something dead simple like spawning a separate app using a URL can be used (e.g., using IHE Invoke Image Display (IID) "http://wiki.ihe.net/index.php/Invoke_Image_Display"), and treating the post-processing app like a "viewer"), may be the expedient approach, given that more sophisticated approaches like DICOM Application Hosting ("http://dicom.nema.org/Medical/Dicom/current/output/chtml/part19/PS3.19.html") have also failed in the marketplace.

Frankly, even getting post-processing applications to generate DICOM output (which can be stored and viewed in the PACS and are properly compliant), rather than being limited to accepting DICOM input, is challenging enough!

David
 Link to this message Neelam Dugar  posted on Friday, November 25, 2016 - 09:45 pm Edit Post Delete Post Print Post
I think pre-processed workflow is required for high volume reporting exams exams--e.g. Chest Xray, Mammography, and limb xray.

Perhaps the images should go from
a. modality to
b. CAD server, and then to
c. PACS.
So that there is "always" the CAD available as a separate series for reading alongside the images. Exam code and description can determine whether an exam goes via CAD or not.

Unless the workflow is efficient "within PACS", CAD is not going to be useful.
 Link to this message Neelam Dugar  posted on Tuesday, December 20, 2016 - 09:18 pm Edit Post Delete Post Print Post
https://www.auntminnie.com/index.aspx?sec=log&URL=http%3a%2f%2fwww.auntminnie.co m%2findex.aspx%3fsec%3dsup%26sub%3dcto%26pag%3ddis%26ItemID%3d116152

Interesting use of CAD-Deep Learning-Artificial Intelligence- whatever you want to call it.

Computer detects abnormality and increases the reporting priority on the worklist for radiologists to report! What is clear is that CAD needs to be truly integrated to a PACS and reporting workflow if there is any chance of it being useful to patient.
 Link to this message Neelam Dugar  posted on Saturday, July 29, 2017 - 07:08 am Edit Post Delete Post Print Post
http://blogs.bmj.com/bmj/2017/06/19/giles-maskell-the-practice-of-radiology-need s-to-change/

http://www.auntminnieeurope.com/index.aspx?sec=sup&sub=pac&pag=dis&itemId=614631

Has any PACS vendor successfully integrated CAD as a 2nd reader for the following:
CXR shadow detection
Mammography shadow detection
Plain X-ray fracture detection
CT lung nodule detection

This would be really valuable in reducing errors--as it would act as 2nd pair of eyes.
 
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users may post messages here.
Password:
Options: Automatically activate URLs in message
Action: