Blog - PYCAD - Your Medical Imaging Partner

Before and after mosaics of fetoscopy, dermoscopy and light-sheet microscopy video: the top row, stitched with an untuned optical flow model, shows grey patches and broken pieces; the bottom row, from the fine-tuned FloVMos model, is continuous (Liu et al., arXiv:2610.04258, Figs. 3 and 5)

Handheld Endoscope or Microscope Video In, One Wide-Field Tissue Mosaic Out at About Ten Frames per Second From a Modality-Tuned Optical Flow Model

Feed FloVMos video from a handheld endoscope, dermoscope or confocal microscope and it stitches the frames into one growing wide-field mosaic, at about ten frames per second on 1 megapixel frames in the authors’ research code. The trick is fine-tuning an optical flow network per device on synthetic motion, so no hand-labelled flow is needed. Several real test sets are one or two videos, accuracy is a feature distance check, a GPU is required, and the repository linked in the paper returned a 404 when I checked. Liu et al., arXiv:2610.04258.

October 10, 2026

Two panels of cardiac MRI slices with absolute error maps below, each showing the ground truth, a UNETR trained from scratch, and a reconstruction from a pretrained DINOv2 encoder, at 4x acceleration in-domain and 10x acceleration across datasets (Hashmi et al., arXiv:2610.08109, Fig. 3)

Undersampled Cardiac MRI In, Sharper Reconstruction Out From a Frozen Pretrained Vision Encoder With About Half the Trainable Weights

Hand an undersampled cardiac MRI slice to a UNETR-style network whose encoder is a frozen pretrained CLIP, BiomedCLIP or DINOv2 model, and the reconstruction beats the same network trained from scratch, with about 49 percent fewer trainable parameters. The gap holds with 5 percent of the training data and across two challenge datasets. It is a 2D study on public CMRxRecon data scored with SSIM and PSNR only, with no reader study, and the linked GitHub repository held only a README when I checked. Hashmi et al., arXiv:2610.08109.

October 10, 2026

Three rows of lung CT crops, a sagittal slice and two zooms, comparing the ground-truth CT, X2CT, the diffusion prior alone, DAPS, and PhyDiCT reconstructed from eight X-ray projections (Dai et al., arXiv:2610.09253, Fig. 2)

Eight X-ray Projections In, a Generated 3D Lung CT Volume Out With No Training but About 22 Minutes of GPU Time

Give it eight simulated X-ray projections around the chest and it steers a frozen, text-conditioned lung CT diffusion model until its own rendered X-rays match, returning a 3D volume with no paired training. It takes about 22 minutes per volume on an RTX 6000, is built for lung CT only, and was tested on X-rays simulated from CT-RATE scans, so the output is a plausible estimate, not a measurement. Apache 2.0 code. Dai et al., arXiv:2610.09253.

October 9, 2026

Rows of coronary artery cases: two segmented angiogram input masks, the CT 3D artery label, a fragmented AutoCAR voxel reconstruction, and the connected PPCAR-Net reconstruction (Ren et al., arXiv:2610.09383, Fig. 4)

Two Coronary Angiogram Masks In, One Connected 3D Artery Tree With Radii Out in About a Tenth of a Second

Give it one to seven segmented coronary angiogram views and their gantry angles, and it returns one connected 3D artery tree with a radius along every branch in about 0.12 s on an RTX 4090, with no point matching between views. Trained and scored on simulated masks from CT. MIT code and checkpoints public. Ren et al., arXiv:2610.09383.

October 9, 2026

Abdominal CT slice with noise and three inserted lesions in circles, shown unfiltered, after Gaussian smoothing scored worse by RIQE, and after bilateral filtering scored best by RIQE while the 4 mm lesion fades (Mattiussi, arXiv:2610.00384, Fig. 4)

CT Slice In, No-Reference Quality Score Out That Catches Dose Noise but Should Not Pick Your Denoiser

Give it one CT slice and it returns a no-reference quality score fitted on public full-dose scans. It flags simulated dose reduction, added noise and blur, but it prefers an edge-preserving filter that wipes out most of a small lesion. MIT code, CC BY model. Mattiussi, arXiv:2610.00384.

October 8, 2026

Sagittal and coronal brain T1 slices with a tumor region and a healthy test region blacked out, the 2D baseline fill showing horizontal slice streaks, and the 2.5D fill continuous across slices (Hadzic et al., arXiv:2610.08983, Fig. 2)

Brain T1 Plus Tumor Mask In, Slice-Consistent Healthy Tissue Fill Out for Registration and Parcellation

Give it a brain T1 and a mask over the tumor, and it paints in healthy-looking tissue that stays continuous across slices, so atlas registration and parcellation can run on a filled copy. Code and checkpoint public. Hadzic et al., arXiv:2610.08983.

October 8, 2026

PYCAD BLOG

Insights on Medical Imaging, AI & DICOM

Handheld Endoscope or Microscope Video In, One Wide-Field Tissue Mosaic Out at About Ten Frames per Second From a Modality-Tuned Optical Flow Model

Undersampled Cardiac MRI In, Sharper Reconstruction Out From a Frozen Pretrained Vision Encoder With About Half the Trainable Weights

Eight X-ray Projections In, a Generated 3D Lung CT Volume Out With No Training but About 22 Minutes of GPU Time

Two Coronary Angiogram Masks In, One Connected 3D Artery Tree With Radii Out in About a Tenth of a Second

CT Slice In, No-Reference Quality Score Out That Catches Dose Noise but Should Not Pick Your Denoiser

Brain T1 Plus Tumor Mask In, Slice-Consistent Healthy Tissue Fill Out for Registration and Parcellation

Company

Our Work

Get in Touch