Cyst Identification and Characterization


Characterize a cyst by clicking on it in PACS with a function to import a textual description directly into the report. 




Non-Interpretive, Reading Room

Define-AI ID



Reading Room Panel

Panel Chair

Joe Cavallo

Panel Reviewers

Alexander J. Towbin, Melissa Davis


Creative Commons 4.0

Status Public Comment 

Clinical Implementation

Value Proposition

The accurate characterization of a cystic lesion on multidetector computed tomography examinations can often be a time-consuming task, particularly when comparing the same lesion on multiple prior examinations or event on different imaging modalities. 

A PACS-centered algorithm which can measure and characterize a lesion and generate a text output for import into the radiology report can increase the efficiency of lesion characterizations. Additionally, standardized descriptive text inserted into structured radiology reports will enhance reporting consistency and possibly decrease interobserver variability. 


A radiologist is reading a CT of the abdomen and pelvis examination and notes a cystic lesion in the right kidney. The radiologist notes that the lesion appears similar when compared to the most recent CT abdomen and pelvis examination. The user manually selects the target lesion. The lesion is then input into a segmentation algorithm which generates a 2D/3D selection boundary. The dimensions of the lesion are obtained by the program/tool. A textual output is generated automatically describing the size and Hounsfield Unit (HU) value of the lesion. The use can then make a selection to either import to report or edit the text for import. 

The radiologist then makes the rest of the interpretation and signs the report. 

Workflow Description

The proposed algorithm will function with the following sequence workflow:

1. The user selects a PACS-integrated button labeled "Cystic Lesion Characterization" to trigger the input to the segmentation algorithm for the cystic lesion on CT exam. 

2. The user selects the center of the cystic lesion and the algorithm employs a segmentation algorithm to define the lesion boundaries and generate a 2D/3D boundary. 

3. Segmentation data is then correlated with the voxel data to generate the anteroposterior, transverse, craniocaudal dimensions and volume. 

4. A language processing application is used to generate the textual description of the segmentation analysis. An "import to report" notification appears and allows the user to import the description text into the report. The imported text can be manually edits by the radiologist, if needed. 

Considerations for Dataset Development

CT-Abdomen/Pelvis (+/- contrast) All examinations where at least one cystic lesion is present. 
CT-Abdomen (+/- contrast) All examinations where at least one cystic lesion is present. 
CT-Pelvis (+/- contrast)

All examinations where at least one cystic lesion is present. 

Technical Specifications


Segmented image of lesion based on user selection


Segmentation & Characterization of cystic lesion



Data Type

Voxel data from DICOM


Computed tomography

Body Region

Abdomen and/or pelvis

Anatomic Focus

Intra-abdominal structures


Primary Outputs

Lesion Descriptor

Size (mm)
Lesion Descriptor Hounsfield Unit
Lesion Location Organ of Origin

Lesion Measurement

Definition Maximum diameter of the segmented lesion on axial image

Data Type Continuous Variable

Value Set N/A

Units mm

Definition Maximum diameter perpendicular to maximum diameter of the segmented lesion on axial image

Data Type Continuous Variable

Value Set N/A

Units mm

Definition Maximum craniocaudal diameter

Data Type Continuous Variable

Value Set N/A

Units mm

Hounsfield Unit Measurement

Definition Mean Hounsfield Unit measurement of cystic lesion

Data Type Continuous numerical value

Value Set  -100 to 1000

Units HU

Lesion Location

Definition Organ of origin for the cyst in question

Data Type String

Value Set Abdomen/Pelvis Organs (automatically classified or selected by radiologist)

Units  N/A

Secondary Outputs 

Transmittal of data to Dictation software

Future Development Ideas

1. Lesion characterization could be used for MRI examinations with which to further classify lesions. 

2. Similar workflow could be used to detect and characterize solid lesions in specific organs. 

3. Future algorithms can use AI to identify the origin of the cyst and further characterize it based on morphology, providing automated follow up guidance based on the risk level of the lesion in questions. 

4. Future algorithm can consider a size threshold above which a cyst could be automatically detected and characterized.