Mammography CAD Workflow - Brief Proposal: Difference between revisions

Revision as of 10:41, 31 August 2007

1. Proposed Profile: Mammography CAD Workflow

Proposal Editor: Carolyn Reynolds
Date: N/A (Wiki keeps history)
Version: N/A (Wiki keeps history)
Domain: Radiology

2. The Problem

CAD processing of Full Field Digital Mammography images is a common practice in the US and is deployed in many parts of the world. Users and vendors have examined applying existing IHE profiles to mammography CAD workflow, such as Post Processing Workflow, without success. In addition to the necessary, enabling actors simply not existing in the marketplace, users appear to desire an on-demand/push model for data transfers, rather than the scheduled/pull model that exists within PPWF.

Although the currently practiced on-demand/push model is preferred for data transfer, it is not without its workflow issues and limitations. Analyzing common workflow issues can easily lead to use cases that could be addressed though an IHE profile.

Below are just a few of the issues the proposed profile could attempt to address.

CAD systems do not know precisely when to start and stop case level processing. Common practice uses a timeout strategy. A timeout strategy delays the receipt of results, which is at issue when a patient is being held for review. The strategy can also not guarantee that it will not miss images when there is any interruption or delay between image acquisitions.

CAD systems do not know for sure which images to include in case level processing. Radiologists often request retakes or additional views after the initial imaging service is completed. In some circumstances, the additional views should be considered along with the original images for case level processing. In other circumstances (such as retakes), there is a preferred view that should replace an original view. Having a mechanism that communicates a logical grouping of objects (such as a modality performed procedure step), and an industry agreed mechanism for indicating case completion would assist mammography CAD workflow.

CAD systems often produce multiple results for the same study. Such is the case when additional views are requested after the initial CAD results are displayed. Should the modality resend all the images for the study so that they can all be included in processing? Should the CAD system be required to retain previously received images until the case is deemed complete? Does the radiologist have a clear indication as to which CAD results to use when multiple results for the same case are available? Can the workstation determine if there is a preferred CAD result to display first within the context of its hanging protocol? Mixing and matching systems that approach these situations differently can leave the user with many challenges.

Although the Mammography Image Profile addresses some Image Display behavior necessary for Mammography CAD support, it is not complete. Some workstations display the CAD results upon default, when initial interpretations must be made before the CAD results are provided as further consideration. Other display issues exist in content and presentation which will be further outlined in the detailed proposal.

Some image display and image managers have issues related to race conditions between image and CAD availability. Radiologists may open a case because the CAD object is present, only to find no images. Some image displays that receive CAD before images arrive have been known to throw away the CAD results. When a radiologist is reviewing images for which a CAD reports arrive, they are unaware that the CAD report is know available and can not display the results without exiting and re-entering the case. Some image managers will mark a study “complete” upon an MPPS “complete” message from the modality. When the CAD report arrives at a later time to a “completed” study, some image displays and image managers disassociate the CAD results from the completed image study.

CAD results are pushed to the image display in practice. Some users do archive CAD results, while some others insist on not archiving results. Using the image manager as the actor for result distribution is inconsistent with users’ desires to not store CAD results. Getting CAD results to multiple workstations presents its own challenges when mixing systems with different capabilities.

3. Key Use Cases

Some use cases can be inferred from the description of the problems above. Other use cases are available, but are omitted in the interest of brevity.

Images are being sent to the CAD system as acquired. The technologist is having a difficult time getting the patient positioned correctly, and takes additional time. The patient moved during the exposure, and the image which shows motion is rejected. Another image is captured and is accepted and sent to the CAD system. The CAD generates two Mammo CAD SR’s – one for the first 3 images and one for the last image. The Radiologist can not view all CAD markings at the same time. There is concern whether or not the CAD results are optimal, given that not all images were considered or case level processing – or were they?

A patient is in for a routine screening exam. The technologist notices a suspicious area during image acquisition and would like the radiologist to review the case before releasing the patient. The patient is being held in the room. The radiologist wants to see the CAD results before releasing the patient. The CAD timeout strategy is set high due to experiencing the situation described in use case #1. The radiologist doesn’t see the CAD results arrive because the study is open and the results can only be accessed by closing and re-opening the study. The patient ties up the room longer than necessary, slowing down the day’s operations.

The CAD arrives at the workstation before the images. The patient name appears on the worklist. The Radiologists selects the patient from the worklist and sees no images.

4. Standards & Systems

Existing standards and mechanisms to consider include:

• DICOM

• Existing IHE profiles, actors, and transactions

Existing systems that could be involved in the problem/solution include: FFDM modalities, Mammography CAD servers, Diagnostic and Technologists’ workstations, PACS, and RIS.

5. Discussion

The Topic “Mammography Acquisition and CAD Workflow” has been separated into two different IHE proposals, although there are several overlapping concerns.

IHE is a great forum to consider solutions to these clinical challenges. The IHE Post Processing Profile likely has many elements and transactions necessary to solve the data and workflow challenges of mammography CAD. However, there are definitely some workflow aspects to post processing which introduce inefficiencies and issues when practiced for mammography. It is anticipated that some operational requirements, specific to the mammography community, may be necessary to solve clinical use cases. For example, requiring modalities to send all images as a case, or requiring CAD to use MPPS messages to determine case content could address case processing concerns.

@@ Line 11: / Line 11: @@
 ==2. The Problem==
-There are common practices in mammography which can have a wide variety of workflow results.  Even when deploying the Scheduled Workflow and Mammography Image Profiles, variances in the way users and systems behave can lead to department inefficiencies, ambiguous data, special cases for automated billing, and less than optimal acquisition and reading environments.
+CAD processing of Full Field Digital Mammography images is a common practice in the US and is deployed in many parts of the world.  Users and vendors have examined applying existing IHE profiles to mammography CAD workflow, such as Post Processing Workflow, without success.   In addition to the necessary, enabling actors simply not existing in the marketplace, users appear to desire an on-demand/push model for data transfers, rather than the scheduled/pull model that exists within PPWF.
-Below are just a few examples of such concerns:
+Although the currently practiced on-demand/push model is preferred for data transfer, it is not without its workflow issues and limitations.  Analyzing common workflow issues can easily lead to use cases that could be addressed though an IHE profile.
+Below are just a few of the issues the proposed profile could attempt to address.
+*	CAD systems do not know precisely when to start and stop case level processing.  Common practice uses a timeout strategy.  A timeout strategy delays the receipt of results, which is at issue when a patient is being held for review.  The strategy can also not guarantee that it will not miss images when there is any interruption or delay between image acquisitions.
-'''CAD Processing and Workflow Race Conditions '''
-*	CAD systems do not know precisely when to start and stop case level processing and exactly what objects should be considered for processing.  Common practice uses a timeout strategy which has its own issues.
+*	CAD systems do not know for sure which images to include in case level processing.  Radiologists often request retakes or additional views after the initial imaging service is completed.  In some circumstances, the additional views should be considered along with the original images for case level processing.  In other circumstances (such as retakes), there is a preferred view that should replace an original view.  Having a mechanism that communicates a logical grouping of objects (such as a modality performed procedure step), and an industry agreed mechanism for indicating case completion would assist mammography CAD workflow.
-*	Race conditions between image and CAD availability can bring about reading inefficiencies.
-*	Additional views complicate case level processing and can result in multiple CAD reports, leaving the workstation and radiologist uncertain as to which CAD report(s) to view.
+*	CAD systems often produce multiple results for the same study.  Such is the case when additional views are requested after the initial CAD results are displayed.  Should the modality resend all the images for the study so that they can all be included in processing?  Should the CAD system be required to retain previously received images until the case is deemed complete?  Does the radiologist have a clear indication as to which CAD results to use when multiple results for the same case are available?  Can the workstation determine if there is a preferred CAD result to display first within the context of its hanging protocol?  Mixing and matching systems that approach these situations differently can leave the user with many challenges.
-'''Determining Exam Completion '''
+*	Although the Mammography Image Profile addresses some Image Display behavior necessary for Mammography CAD support, it is not complete.  Some workstations display the CAD results upon default, when initial interpretations must be made before the CAD results are provided as further consideration.  Other display issues exist in content and presentation which will be further outlined in the detailed proposal.
-*	Because additional views (both on the day of the study and at a later time) are common practice and other objects (annotations, CAD results) can be added to studies at a later time, it is difficult for both users and systems to determine when a patient exam is actually complete.  Many systems have implicit ways of treating exams as complete or open which can conflict with the actual state of the mammography exam.
+*	Some image display and image managers have issues related to race conditions between image and CAD availability.  Radiologists may open a case because the CAD object is present, only to find no images.  Some image displays that receive CAD before images arrive have been known to throw away the CAD results.  When a radiologist is reviewing images for which a CAD reports arrive, they are unaware that the CAD report is know available and can not display the results without exiting and re-entering the case.  Some image managers will mark a study “complete” upon an MPPS “complete” message from the modality.  When the CAD report arrives at a later time to a “completed” study, some image displays and image managers disassociate the CAD results from the completed image study.
+*	CAD results are pushed to the image display in practice.  Some users do archive CAD results, while some others insist on not archiving results.  Using the image manager as the actor for result distribution is inconsistent with users’ desires to not store CAD results.  Getting CAD results to multiple workstations presents its own challenges when mixing systems with different capabilities.
 ==3. Key Use Cases==
@@ Line 38: / Line 40: @@
-.	A patient comes in for a screening mammogram which is converted to a diagnostic exam upon the technologist discovering a lump.  The tech adds an extra view, but no exam type, so the hanging protocol at the workstation is unacceptable.
+*	Images are being sent to the CAD system as acquired.  The technologist is having a difficult time getting the patient positioned correctly, and takes additional time.  The patient moved during the exposure, and the image which shows motion is rejected.  Another image is captured and is accepted and sent to the CAD system.  The CAD generates two Mammo CAD SR’s – one for the first 3 images and one for the last image.  The Radiologist can not view all CAD markings at the same time.  There is concern whether or not the CAD results are optimal, given that not all images were considered or case level processing – or were they?
-.	A patient comes in for follow-up in one breast.  Since she is close to her yearly exam date, the radiologist calls for imaging the other breast while the patient is waiting.  The original room is unavailable, so another machine is used.  The tech generates a new study and accession number.  The right and left breasts do not hang together at the diagnostic workstation.  The following year, the prior images are hung as two prior studies, not as one complete comparison study.
-.	A radiologist views a screening exam and toggles on CAD which shows findings.  The radiologist requests additional images to better demonstrate the abnormality.  Should the tech add the views to the existing order?  Does it depend upon whether or not the patient is still available for imaging that day?  How should the exam type (and billing) change from screening to diagnostic if this is necessary?
-.	The technologist labels a view incorrectly and doesn’t realize it until after the case is completed and sent to PACS.  The view information is corrected at the acquisition modality and the view resent.  The images do not hang properly on the workstations and multiple copies of the same image exist, one which is labeled incorrectly.
+*	A patient is in for a routine screening exam.  The technologist notices a suspicious area during image acquisition and would like the radiologist to review the case before releasing the patient.  The patient is being held in the room.  The radiologist wants to see the CAD results before releasing the patient.  The CAD timeout strategy is set high due to experiencing the situation described in use case #1.  The radiologist doesn’t see the CAD results arrive because the study is open and the results can only be accessed by closing and re-opening the study.  The patient ties up the room longer than necessary, slowing down the day’s operations.
-.	A PACS will not distribute images until a study is complete.  A technologist wants the radiologist to view the images before completing the exam.  The radiologist can not access the images from the local workstation.
+*	The CAD arrives at the workstation before the images.  The patient name appears on the worklist.  The Radiologists selects the patient from the worklist and sees no images.
@@ Line 67: / Line 62: @@
 ==5. Discussion==
-The Topic “Mammography Acquisition and CAD Workflow” is currently a catch-all topic for issues related to the workflow of ordering, acquisition, processing, archiving, display, and billing for mammography procedures.  Upon further investigation, the IHE Mammography committee may further refine the scope of this proposal to keep aligned with the time and resources available to deliver viable solutions to specific use cases.
+The Topic “Mammography Acquisition and CAD Workflow” has been separated into two different IHE proposals, although there are several overlapping concerns.
-IHE is a great forum to consider solutions to these clinical challenges. The challenges with additional views might be solved by defining how to apply more generalized scheduled workflow use cases to mammography practices.  Where existing profiles may not meet mammography departmental needs, it is believed that the IHE venue may be able to identify other solutions or identify gaps in technology that could be handed off to the appropriate resources (i.e. DICOM committees).
-The IHE Post Processing Profile has been discussed as a candidate for mammography CAD workflow.  An effort into understanding why this profile has not been adopted and addressing barriers to its adoption could lead to agreeable solution (or a more viable profile) among the mammography community without waiting for the greater radiology community to adopt the already existing profile.
+IHE is a great forum to consider solutions to these clinical challenges.  The IHE Post Processing Profile likely has many elements and transactions necessary to solve the data and workflow challenges of mammography CAD.  However, there are definitely some workflow aspects to post processing which introduce inefficiencies and issues when practiced for mammography.  It is anticipated that some operational requirements, specific to the mammography community, may be necessary to solve clinical use cases.  For example, requiring modalities to send all images as a case, or requiring CAD to use MPPS messages to determine case content could address case processing concerns.