Dorit Merhof

This work presents a trained deep registration pipeline that deformably aligns a larval brain to a reference template in a single forward pass at high spatial resolution, on volumes that hold several times more voxels than those learned 3D registration is normally reported on, together with the preprocessing and anatomy-anchored evaluation pipeline required to apply it.

State-of-the-art 3D pancreatic tumor segmentation no longer demands massive architectures or pre-training: a tailored 5.86M-parameter model boosts lesion sensitivity by 6.6 points simply by replacing standard pooling with lossless downsampling and linear-deformable attention.

MaLViL slashes memory usage by 83x while achieving state-of-the-art accuracy in medical image segmentation.

Achieving a 0.20% accuracy loss while eliminating BRAM usage could revolutionize the deployment of Vision Transformers on resource-constrained edge devices.

Achieving up to 82% higher energy efficiency while maintaining accuracy, MINT redefines the boundaries of CNN inference on FPGA platforms.

Achieving 80% higher throughput with dynamic precision in ultrasound beamforming could revolutionize real-time imaging applications on FPGA devices.

Merging multiplication and addition in FPGA architectures can boost energy efficiency for image segmentation tasks by up to 9x compared to existing methods.

Achieve state-of-the-art medical image segmentation accuracy with a linear-time transformer decoder that overcomes the limitations of standard linear attention by subtracting complementary attention paths to amplify relevant context.