Development of a Question Answering based conversational system that can provide users information on GI Cancer. User may or may not be a physician. The information asked may be about general symptoms and diagnosis of various types of GI cancers or can be of focused nature based on medical history of a person.
What to look for - some pointers:
Relevant entities and relations that can be extracted from text for building domain knowledge
Entities:
1. Patients: Demographics, medical history, lifestyle factors, symptoms.
2. Tumors: Location, size, stage, type, grade, molecular profile.
3. Tests: Imaging tests (X-ray, CT, MRI, PET), endoscopic procedures
(colonoscopy, EGD), blood tests, biopsies.
4. Biomarkers: Specific proteins or genetic markers associated with GI
cancer.
5. Treatments: Surgery, chemotherapy, radiation therapy, targeted therapy,
clinical trials.
6. Outcomes: Survival rate, disease progression, response to treatment,
quality of life.
Relationships:
1. Patients - have medical history, symptoms, tumors.
2. Tumors - are diagnosed by tests and biopsies.
3. Tests and biopsies- reveal specific biomarkers and tumor characteristics.
4. Tumor -characteristics guide treatment selection.
5. Treatments- have associated risks and benefits.
(Please note, this is just a sample template, to kickstart the system. This is not exhaustive, nor mandatory to follow. Participants have freedom to implement in whatever ways they prefer)
Policy on using LLM:
If needed, participants can use free and open source LLMs or SLMs. Commercial LLMs
Evaluation plan –
We will provide several Questions, against which the participants need to submit the answers generated by their system. The answers will be evaluated against our ground truth answers on the following criteria:
a. Concepts/entities/relationships correctly identified
b. Linguistics correctness and meaningfulness of the answers
c. Consistency in the answers when asked similar question with different paraphrases
d. Confidence in the questions when doubted