Opportunistic screening for low bone density using abdominopelvic computed tomography scans.

Area of Science:

• Radiology
• Artificial Intelligence
• Bone Health

Background:

• Low bone density poses a significant public health challenge, yet current screening guidelines have limitations in reach and adherence.
• Opportunistic screening via abdominal CT could identify undiagnosed cases, especially in populations not meeting traditional screening criteria.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) model for opportunistic screening of low bone density.
• The model aims to utilize abdominopelvic computed tomography (CT) scans (both contrast and non-contrast) for identifying patients who may benefit from further bone health assessment, such as dual-energy X-ray absorptiometry (DXA).

Main Methods:

• A dataset of 6083 contrast-enhanced CT exams paired with DXA scans was used for model training and testing.
• A fusion AI pipeline integrated imaging data (coronal and axial planes) with patient demographics (age, gender, body dimensions).
• The model was further validated prospectively on 344 contrast-enhanced and non-contrast-enhanced CT studies.

Main Results:

• The fusion AI model achieved an area under the receiver operating characteristic curve (AUROC) of 0.86, outperforming models based on demographics or single imaging views alone.
• In prospective testing, the model demonstrated high accuracy, with a 0% false positive rate for non-contrast studies and successful detection of positive cases.
• The model identified approximately 30% of patients with low bone mass in the prospective cohort, highlighting its potential for opportunistic screening.

Conclusions:

• A fusion AI model combining CT imaging and electronic medical record data offers robust diagnostic performance for opportunistic low bone density screening.
• This AI approach can enhance bone health risk assessment cost-effectively, potentially improving patient outcomes.
• Further development is ongoing to address challenges such as the model's performance with metallic implants.