An Image Analysis Framework for Objective Color Interpretation of Seized Drug Tests

Color testing continues to be reported as the most common seized drug screening method because of the advantages of low cost, rapid results, and simple visual interpretation. However, limitations from human factors considerations, interferents, and manual interpretation and recording of results often lead to opportunities for user error and unreliable results. Advancements in digitization and digitalization could be used address these limitations.

Forensic disciplines have implemented digital techniques to support investigations and data collection (e.g., electronic tracking systems, automated data acquisition, reference databases, and digital evidence processing). Although there have been significant technological advancements since the early implementation of color tests for seized drugs, many of these methods have not been fully explored or integrated into color test workflows today. However, these tests are advantageously amenable to digital improvements using imaging. Exploiting the accessibility of digital cameras as an objective detector hinges on a software platform capable of advanced computing.

Methodology critical for interpreting colorimetric results in a simple and user-friendly format should include image processing, automated selection, and defined threshold values. Resources such as ImageJ or other commercially available algorithm-writing software are best suited for research purposes rather than forensic science service provider (FSSP) implementation due to the lack of automation and number of manual steps that can be time-consuming when processing many images. Alternatively, developing custom software for needs such as throughput, automation, and objective analysis beyond research-grade tools requires complex processing algorithms, advanced computer programming knowledge, and expertise in computer vision and learning.

This proposed work will use an in-house software platform for an agnostic framework for image processing and objective color interpretation. Phase 1 of this proposed research will include the generation of images for each routinely used color test and define optimal analysis parameters for each. Subsequently, Phase 2 will include machine learning algorithms to improve automation, user interface and output design, and a software platform pilot with collaborating FSSPs to support evidence processing workflows. Ultimately, this work aims to leverage a digital solution to produce more reliable results by decreasing variabilities from human factors considerations that may impact interpretation and subjectivity, allowing for the storage of data for traceability, recall, or review, while requiring minimal changes to current FSSP workflows (e.g., no additional instrumentation or hardware) for more streamlined transition and implementation. CA/NCF