3 Matching Annotations
  1. Mar 2024
    1. Millions of Patient Records at Risk: The Perils of Legacy Protocols

      Sina Yazdanmehr | Senior IT Security Consultant, Aplite GmbH Ibrahim Akkulak | Senior IT Security Consultant, Aplite GmbH Date: Wednesday, December 6, 2023

      Abstract

      Currently, a concerning situation is unfolding online: a large amount of personal information and medical records belonging to patients is scattered across the internet. Our internet-wide research on DICOM, the decade-old standard protocol for medical imaging, has revealed a distressing fact – Many medical institutions have unintentionally made the private data and medical histories of millions of patients accessible to the vast realm of the internet.

      Medical imaging encompasses a range of techniques such as X-Rays, CT scans, and MRIs, used to visualize internal body structures, with DICOM serving as the standard protocol for storing and transmitting these images. The security problems with DICOM are connected to using legacy protocols on the internet as industries strive to align with the transition towards Cloud-based solutions.

      This talk will explain the security shortcomings of DICOM when it is exposed online and provide insights from our internet-wide research. We'll show how hackers can easily find, access, and exploit the exposed DICOM endpoints, extract all patients' data, and even alter medical records. Additionally, we'll explain how we were able to bypass DICOM security controls by gathering information from the statements provided by vendors and service providers regarding their adherence to DICOM standards.

      We'll conclude by providing practical recommendations for medical institutions, healthcare providers, and medical engineers to mitigate these security issues and safeguard patients' data.

  2. Oct 2023
    1. Neural operators are guaranteed to be discretization invariant, meaning that they can work on any discretization of inputs and converge to a limit upon mesh refinement. Once neural operators are trained, they can be evaluated at any resolution without the need for re-training. In contrast, the performance of standard neural networks can degrade when data resolution during deployment changes from model training.

      Look this up: anyone familiar with this? sounds complicated but very promising for domains with a large range of resolutions (medical-imaging, wildfire-management)

  3. Mar 2021