This is the official repository for the code analysis of the paper "Improving Basal Cell Carcinoma Diagnosis with Real-Time AI Assistance for LC-OCT: a multi-centric retrospective study".
Retrive the code by cloning this repository
For full reproducibility, this code comes with a docker environment that can be installed following those steps:
- go to
cd ./code-analysis-bcc-classification-results/ - build the docker image and install all poetry dependencies using make
make install - launch jupyter by typing
make notebook - then follow the notebooks url to access the jupyter environment in which you can run the code
If you do not want to use the docker container, you can simply create a corresponding poetry environment using the pyproject.toml file.
The data folder contains three csv files which contain all the results of our study.
data/simple_diags_CT.csv
This csv (";" separated) is a mapper from all the diagnosis to normalized categories
- 'case_id': unique identifier of the corresponding lesion
- 'histological_diagnosis': diagnosis as written by the pathologist
- 'category': normalized category associated to the histological diagnosis
data/BCC_Quiz_answers_anon.csv
This csv ("," separated) contains all the answer to each question by each doctor. The columns should be understood as follow:
- 'user': unique id of the doctor
- 'case_uuid': unique id of the the studied case
- 'is_diagnostic_bcc': histological result, True for BCC else False
- 'ai_assistance_present': was the AI assistance present for this question during the LC-OCT phase
- 'body_location': body location of the lesion
- 'clinical_phase_answer': answer to the "Is it BCC?" question at the clinical imagery phase
- 'clinical_phase_trust_score': confidence score given to the diagnosis at the clinical imagery phase
- 'clinical_phase_elapsed_time': time in seconds for answering at the clinical imagery phase
- 'clinical_phase_remark': free comment of the doctor on this specific case
- 'lcoct_phase_answer': answer to the "Is it BCC?" question at the lc-oct imagery phase
- 'lcoct_phase_trust_score': confidence score given to the diagnosis at the lc-oct imagery phase
- 'lcoct_phase_elapsed_time': time in seconds for answering at the lc-oct imagery phase
- 'lcoct_phase_remark': free comment of the doctor on this specific case at the lc-oct imagery phase
- 'answered_order': number indicating after how many previous questions this case was answered
- 'date': date at which the question was answered
- 'img_name': duplicate of case_uuid
- 'max_moving_avg_24': max moving average over 24 consecutive frames of the AI model on this LC-OCT images
- 'all_scores': frame level scores of the AI model
- 'user_type': level of expertise of the doctor
data/BCC_Quiz_answers_anon.csv
This csv file ("," separated) contains case level (not answer level) ai predictions scores.
- 'station_name': identifier of the device used by doctor when acquiring this case
- 'patient_id': unique identifier of the patient
- 'study_id': unique identifier of the exam number
- 'series_nb': number corresponding to this image in the study
- 'img_name': unique identifier of the image inside the exam number
- 'img_type': original img type (2D video or 3D)
- 'patient_sex': sex of the patient as written by the doctor during exam
- 'body part': body part as entered by the doctor during exam
- 'max_moving_average': maximum moving average score of 24 consecutive frames of the AI model
- 'all_scores': all scores of the AI model at the frame level
- 'case_uuid': duplicate of image_name
- 'is_diagnostic_bcc': histological result, True for BCC else False
The function used to compute all the statistic and make the plots of the study are located in the src folder.
src/CI_utils.pycontains function to compute sensitivity and specificity confidence intervalssrc/plot_utils.pycontains function to make the plots of the publicationnotebooks/Compute_statistics.ipynbcontains all the code to reproduce the statistical results of the paper.
The notebook should run in a few seconds (1 min max) on a modern laptop.