DeepPhe Software
The DeepPhe software tools use a combination of natural language processing, machine learning, visual analytics,
a rich ontology and ontology-based summarization to extract longitudinal histories from cancer clinical notes.
DeepPhe software can provide summary phenotype and patient information as well as facilitate retrospective cancer cohort research.
Built on the widely-used Apache cTAKES clinical NLP platform, the DeepPhe NLP Summarizer can extract
tumor characteristics, biomarkers, treatments, and comorbidity information, as well as classify a clinical note’s episode in oncology care.
DeepPhe software products
- DeepPhe Translational - for translational studies and research cohorts.
- DeepPhe-CR - for cancer registry applications and registrar workflows.
Though both software products are used for Cancer Phenotype extraction, capabilities and execution are extremely different.
DeepPhe Translational
DeepPhe Translational is the DeepPhe software suite for translational studies and research cohort discovery. It provides extensive information useful for translational cancer research. Results from processing clinical documents can be stored in a SQLite database that can be utilized (in conjunction with OMOP tables) by the DeepPhe Visualizer.
The DeepPhe Translational NLP Patient Summarizer is a highly-modifiable pipeline that can extract values for cancer and tumor attributes such as:
Diagnosis, Location, TNM Stage, Summary Stage, and Grade.
The DeepPhe Translational NLP Patient Summarizer can also identify cancer-related concepts such as:
Biomarkers, Medications, Procedures, Comorbidities, and Genes.
A DeepPhe Translational pipeline can be configured to report minimum, maximum and confidence-thresholded discoveries,
as well as output selected discoveries in a variety of file formats such as CSV tables and JSON files.
The pipeline can be resized to accommodate smaller projects, and custom engines can be added to perform additional operations,
such as parsing institution-specific metadata in document headers or writing extracted information to new output formats.
Use of the default pipeline is simple, but the default pipeline may not be best for your data or use case.
Customization is highly recommended. For work on refined workflows or custom operations, contact the DeepPhe team.
Supported by the National Cancer Institute’s Information Technology for Cancer Research (ITCR) initiative (Grant #U24CA248010).
DeepPhe Visualizer
DeepPhe results that are stored in a SQLite database such as those created by DeepPhe Translational can be utilized with the DeepPhe Visualizer, which provides a web-based interface to data, supporting review of both cohorts and individual patient records.
See the DeepPhe Visualizer 2 User Guide for installation and usage documentation.
Documentation: DeepPhe Data API for the API layer that serves the database read by DeepPhe Visualizer.
Supported by the National Cancer Institute’s Information Technology for Cancer Research (ITCR) initiative (Grant #U24CA248010).
DeepPhe Related Software
The DeepPhe Project includes several software products that are not currently integrated with the DeepPhe Translational software.
They are useful for cancer-related tasks and can be dowloaded and run separately.
At this time they are kept in the HealthNLP .org repository,
along with other code projects by DeepPhe team members.
The capabilities of these products will be added to DeepPhe Translational in a future release.
- HNLP-TimeNorm
- Provides models for finding natural language expressions of dates and times and converting them to a normalized form.
- LGT-SACT
- Extracts and normalizes temporal information from clinical notes using fine-tuned LLMs. Specifically, Systemic Anticancer Therapy (SACT) Timelines.
- ClinGen
- Extracts clinical genomic information from clinical notes using LLMs.
DeepPhe-CR
The DeepPhe-CR Final release is the project’s final software release for cancer registry applications. DeepPhe-CR is intended for cancer registry workflows, registrar case abstraction, and integration with registry systems.
DeepPhe-CR provides REST APIs
for submissions of documents to the DeepPhe-CR pipeline and return of extracted results,
with the goal of providing DeepPhe services to tools for cancer patient data abstraction.
Developed in support of cancer registries from the NCI
Surveillance, Epidemiology, and End Results (SEER) Program,
DeepPhe-CR provides web services suitable for integration into existing registry
case abstraction tools.
DeepPhe-CR tools are provided as a suite of Docker containers supporting ease
of installation and operation.
DeepPhe-CR has been developed in partnership with the
Kentucky Cancer Registry.
DeepPhe-CR is not a web or REST-based version of DeepPhe Translational.
DeepPhe-CR has different capabilities, a different ontology, and different processors.
DeepPhe-CR provides one and only one ICD-O code for a primary cancer’s morphology, topography, laterality, and grade.
This behavior is very different from that of DeepPhe Translational, which reports 0-n values for numerous attributes of multiple cancers and tumors.
There is a DeepPhe-CR wiki with information on installation and use.
A collaboration with the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) Program (Grant #UH3CA243120).