MonMIND 2025
Montreal Medical Imaging Networking Day

MonMIND 2025

(May 2nd, 2025, Polytechnique Montreal)

What: Montreal Medical Imaging Workshop
When: May 2, 2025, 9:30am-4:30pm
Where: Polytechnique Montreal, Galerie Rolland (B-600.16), 500 Chem. de Polytechnique, Montréal, QC H3T 0A3

About MonMIND: Montreal Medical Imaging Networking Day, MonMIND serves as a dynamic hub for students and trainees to showcase their research and engage in high-level scientific exchange with peers and experts. Its primary goal is to spark interdisciplinary collaborations that bridge the gap between technical imaging innovation and real-world clinical applications.

Lunch and Coffee will be provided

Schedule (2025)

09:30 - introduction + coffee

10:00 - Student Session #1 (4 x 15min presentations)

1. Castillo, Joshua - iSurgARy: A mobile augmented reality solution for ventriculostomy in resource-limited settings

2. Javadi, Soroush - Machine Learning-Assisted Guidance for Novice Sonographers in Kidney Imaging

3. Gheflati, Behnaz - Leveraging deep learning for nonlinear shape representation in anatomically parameterized statistical shape models

4. Hashemibakhtiar, Pejman - 2D/3D Reconstruction of Distal Tibiofibular Joint from Ankle Biplanar Radiographs by a Pipeline of Deep Networks 

11:00 - Break

11:15 - Keynote: Steve Pizer - Object Correspondence for Shape Statistics via Skeletal Geometry

12:15 - Lunch

13:15 - Student Session #2 (4 x 15min presentations)

1. Gaillochet, Mélanie - Weakly supervised prompt learning with box-based constraints for medical image segmentation

2. Popa, Bianca - Multimodal segmentation and axial simulation of traumatic spinal cord injuries using deep learning

3. Dargahi, Sedigheh - Susceptibility Distortion Correction of Diffusion MRI with a single Phase-Encoding Direction

4. Fedorenko, Dmytro - Using GANs to enhance confocal microscopy images

14:15 - Break

14:30 - Student Session #3 (4 x 15min presentations)

1. Saremi, Parham - Conditional Diffusion Models Are Medical Image Classifiers That Provide Explainability And Uncertainty For Free

2. Beizaee, Farzad - Masked Diffusion for Unsupervised Brain Anomaly Detection

3. Shakeri, Fereshteh - Zero-shot and few-shot adaptation of Medical Vision Language Models

4. Salari, Soorena - CABLD: Contrast-Agnostic Brain Landmark Detection with Consistency-Based Regularization

15:30 - Social Activity

16:30 - End of event

Keynote 2025 – Stephen Pizer, PhD, Kenan Professor of Computer Science, UNC

Object Correspondence for Shape Statistics via Skeletal Geometry

By: Stephen Pizer, PhD, Kenan Professor of Computer Science, UNC

Abstract

Effective statistical analysis, such as classification or hypothesis testing, or segmentation of image objects depends on locational correspondence between the objects in the population. This correspondence can be accomplished by producing the statistics, i.e., the object features, by starting from a common base object and deforming it in a way that richly reflects the object geometry. I present a method whereby 2nd and 1st order object features are derived from a skeletal shape representation characterizing the geometry of the object interior, where throughout the whole deformation sequence the deforming representation is matched to a variety of critical object loci on the deforming object. A central aspect is the use of a fitted coordinate frame at each location to provide local coordinates in the object interior and boundary, thereby avoiding dependence on object pre-alignment. Also of importance is the understanding that coordinate frames and direction vectors require statistics on the abstract spheres on which they reside. A comparison of this approach with the more classical deformation approach and with an earlier skeletal approach will be reported, showing notable classification performance improvement over the alternatives, thus suggesting that this object representation should be used for diagnosis, hypothesis testing, and segmentation of anatomic objects affected by various disorders. The comparison measures the success in classifying a baby’s hippocampus as to whether he or she will develop autism. Finally, a discussion of a variant of this skeletal representation due to Taheri that guarantees that object population statistics such as means themselves are geometrically valid will be sketched.