2021 Fiscal Year Research-status Report
Explainable Artificial Intelligence for Medical Applications
| Project/Area Number |
21K17764
|
|---|
| Research Institution | Osaka University |
|---|
Principal Investigator |
李 良知 大阪大学, データビリティフロンティア機構, 特任助教(常勤) (10875545)
|
|---|
| Project Period (FY) |
2021-04-01 – 2023-03-31
|
| Keywords | Explainable AI / Computer Vision / Medical Images / Deep Learning / Image Classification / Visual Explanation / Computer-aided Diagnosis / Trustable AI |
|---|
| Outline of Annual Research Achievements |
To fully enable trustworthy AI for medicine and healthcare, this project aims to design an explainable AI model that can give diagnosis results along with precise and bifunctional visual explanations to support its decisions. In FY 2021, I mainly studied the following sub-topics towards the goal.
1. A self-attention-based classifier that has the ability to conduct intrinsically-explainable inference.
2. A loss function for controlling the size of explanations. I design a dedicated loss named explanation loss, which is used to control the overall explanation size, region number, etc., of the visual explanations.
3. Collaborating sub-networks to output positive and negative explanations simultaneously.
The results are presented in the IEEE ICCV 2021 conference.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
I planned to study the following sub-topic in FY 2021 and finished it as scheduled.
1. A self-attention-based classifier.
2. A loss function for controlling the size of explanations.
3. Collaborating sub-networks.
In addition, I also get some results from other two sub-topics.
1. Tests on medical diagnosis, including glaucoma, artery hardening, etc.
|
| Strategy for Future Research Activity |
1. I plan to extend the work to an explainable regressor, which can deal with regression tasks like disease severity estimation
2. This project will also involve an evaluation of the explanation quality. The evaluation will be performed by medical professionals and will serve as feedback to improve the design of explanation loss.
3. I plan to include an evaluation of machine teaching, in which the proposed method will be used to teach humans using the generated visual explanations. Machine teaching is a commonly-used evaluation in the XAI field and the teaching effects will be measured by the test scores of the participants.
|
| Causes of Carryover |
Reasons for Incurring Amount to be Used Next Fiscal Year: Many international conferences have been cancelled due to the pandemic. So the fee for
travel and registration are not used.
Usage Plan: In the next year, the funding will be used for buying the workstation and other hardwares to build the prototype system, and the registration fee of the conferences that will be held online next year.
|
