Apparatus for and method of de-identification of medical images
Inventors
Chunduru, Abhijith • SAHA, Dibakar • Kaluva, Krishna Chaitanya • Vaidya, Suthirth
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12204681-B1
Publication Date
2025-01-21
Expiration Date
Abstract
An apparatus and method for de-identification of medical images including at least a processor and a memory communicatively connected to the at least a processor, the memory containing instructions configuring the at least a processor to receive a series of images, comprising metadata, and a plurality of image slices; select a sampling strategy as a function of the metadata, wherein the sampling strategy identifies a subset of image slices of the plurality of image slices to sample; apply a text classifier to the subset of image slices, wherein the text classifier is configured to identify a presence of textual information on the subset of image slices; determine at least one relationship between the subset of images slices including providing an output as a function of the presence of the textual information, the metadata, and a position of each image slice.
Core Innovation
The disclosed invention is a de-identification system for a series of images comprising metadata and a plurality of image slices, such as image slices in a medical image series. The system selects a sampling strategy as a function of the metadata so that only a subset of image slices is sampled, thereby reducing computational cost.
For the sampled subset, the system applies a text classifier configured to identify a presence of textual information on the subset of image slices. Using the presence of the textual information, the metadata, and a position of each image slice in the series, the system determines relationships and provides an output that indicates whether to mask all image slices of the plurality of image slices.
Based on the detected textual information and the output, the system masks one or more image slices of the plurality of image slices. Masking decisions include masking all slices versus masking one or more slices, and may blank, remove, modify, or redact textual content, optionally using masking thresholds and information-class-based PII categorization.
Claims Coverage
The partial content identifies two independent claims, an apparatus claim and a method claim, each centered on the same core inventive workflow: metadata-dependent sampling of a subset of image slices, text-classifier detection on the sampled subset, determining relationships using detected text plus metadata plus slice position, and using an output to decide whether to mask all slices or only one or more slices. The independent claims contain a total of several dependent claim refinements in the families summarized in the provided material.
Metadata-dependent slice sampling for text analysis
Selects a sampling strategy as a function of the metadata, wherein the sampling strategy identifies a subset of image slices of the plurality of image slices to sample.
Text classifier for presence of textual information on sampled slices
Applies a text classifier to the subset of image slices identified by the sampling strategy, wherein the text classifier is configured to identify a presence of textual information on the subset of image slices.
Relationship-determined masking decision from detected text, metadata, and slice position
Determines at least one relationship between the subset of image slices by providing an output as a function of the presence of the textual information, the metadata, and a position of each image slice of the subset of image slices in the series of image slices, wherein the output indicates whether to mask all image slices of the plurality of image slices.
Mask one or more image slices based on textual information and masking decision
Masks one or more image slices of the plurality of image slices as a function of the textual information and at least the output.
Across the independent apparatus and method claims, the inventive combination is metadata-dependent sampling to sample a subset of image slices, using a text classifier on the sampled subset to identify presence of textual information, determining at least one relationship and producing an output using detected textual information, metadata, and slice position, and masking one or more image slices based on the textual information and the output.
Stated Advantages
Reduces compute cost by analyzing only a subset of image slices via a metadata-dependent sampling strategy.
Documented Applications
De-identification of medical images in image series such as DICOM stacks using slice-level masking based on detected textual/PII content.
Series-level vs slice-level masking workflows for CT, MR, and PET image series.
Combining or merging masking decisions with other de-identification processes, including skull stripping.
Interested in licensing this patent?