Ultra-Low Dose CT for Screening Applications

Why screening needs a better modality

Screening is "the process of using tests on a large scale to identify the presence of disease in apparently healthy people" . To work at population scale it must be low-cost and safe. Yet in half a century the USA has fielded only a handful of large-scale, reimbursed imaging-based screening programs — mammography (1976), osteoporosis DEXA (1998), and low-dose CT for lung cancer (2015).

The bottleneck is that every existing modality trades strengths against weaknesses. CT offers rich 3D information at low cost and high availability, but its ionizing radiation has kept it off the screening table for most applications.

The solution: AI-enhanced ultra-low dose CT

The proposal is to bring any existing, low-cost CT scanner down to the radiation level of a 2D X-ray — turning the "radiation" weakness into a "negligible" — and thereby enable entirely new screening applications.

The motivation is concrete: CT screening for lung cancer reduces mortality by ~20%, and annual low-dose lung screening CT is reimbursed for at-risk populations in the US, amounting to some 17 million scans per year. But "low dose" is still ~1.5 mSv — the equivalent of about 15 chest X-rays. Following the ALARA principle (As Low As Reasonably Achievable), the goal is to close that gap.

Building a dataset that doesn't exist

Supervised denoising needs paired ultra-low-dose and normal-dose scans of the same patient — a dataset that simply did not exist. There are two ways to obtain one:

• Retrospective — add synthetic noise to existing normal-dose scans. One scan per patient, aligned by design, no consent required — but the noise is simulated.
• Prospective — acquire real pairs of ULD and normal-dose scans. Two scans per patient, requiring informed consent and non-rigid registration — but the noise is real.

Real noise matters: models trained on genuinely acquired ULD noise achieve significantly higher Peak-SNR, higher structural similarity (SSIM), and higher S3 sharpness (all statistically significant) than those trained on synthetic noise.

Because a patient never holds exactly the same full inspiration twice, each ULD scan is brought into alignment with its normal-dose ground truth through affine + deformable registration (Elastix).

The denoising engine: perceptual loss to PeTeL

The denoising network is trained with a perceptual loss — comparing denoised and ground-truth patches inside a learned embedding space rather than pixel-by-pixel. The lab's approach evolved from a convolutional perceptual loss (ISBI 2019) to the Perceptual Transformer Loss (PeTeL) built on a 3-D MaxViT backbone combining block and grid self-attention, which achieves superior quantitative and qualitative denoising (ISBI 2024).

One scan, four applications

A single AI-denoised ultra-low dose acquisition supports four distinct clinical reads. The technology has obtained regulatory clearance from the Israeli Ministry of Health and is in clinical operation at Sheba Medical Center — offering lung cancer screening at among the world's lowest radiation doses.

Publications

On the algorithmic side, the lab introduced a perceptual transformer loss for the neural denoising of ultra-low dose CT (IEEE ISBI, 2024) and a novel artificial-intelligence-based denoising method for ultra-low dose CT used for lung cancer screening (Academic Radiology, 2023). These methods have been validated across clinical tasks: denoised ultra-low-dose chest CT to assess pneumonia in immunocompromised individuals (Radiology: Cardiothoracic Imaging, 2025); denoising of ultra-low dose chest CT for coronary artery calcium score evaluation (Clinical Radiology, 2022); ultra-low dose chest CT with denoising for lung nodule detection (IMAJ, 2021); and liver attenuation assessment in reduced-radiation chest computed tomography (J. Computer Assisted Tomography, 2022).