01

1. Start with the unit being counted

The 2024 study reports 923 usable clips, with 184 used for testing. Rayong reports 218 participants. These are different units: a clip is not necessarily one unique person, so the counts must not be added into a total number of tested patients.

The same principle applies to activity photographs or cumulative counts in slides. Without a unit definition, counting method, and handling of duplicates, such a figure should not substitute for a research sample size.

Source: 2024 paper, pp. 123–124, 128; Rayong abstract, pp. 1–2

02

2. Separate learning data from test data

In 2024, 739 clips were used for training and 184 for testing. Strong performance on learning data does not answer how well a model handles new data. The principal reported result is therefore 177/184, or 96.2%, on the three-category test task.

For a later model, reviewers should establish how test data relates to training data and whether separation is by participant or site. These are methodological questions, not an assertion that the earlier study had data leakage.

Source: 2024 paper, pp. 128–131

03

3. Overall performance does not represent every subgroup

Overall accuracy of 96.2% can coexist with only 12/17 correct Mildly Reduced EF clips, approximately 71%, because group sizes differ. Likewise, Rayong’s 93.6% overall result coexists with 27.3% for fair LV. The intermediate-category result is part of the main explanation, not an optional detail.

A very high percentage within a category still depends on sample size and setting. Without sufficient data, readers should not invent confidence intervals or relabel accuracy as sensitivity or specificity.

Source: 2024 paper, pp. 124, 130–132; CNF 2026, p. 59

04

4. Compare like questions under like conditions

The 68.5% five-category and 96.2% three-category results in the 2024 paper measure different levels of detail. Consolidation changes what counts as an error, so the percentages cannot be treated as a context-free comparison of processing quality.

Similarly, the 2024 figure of 96.2% and Rayong’s 93.6% should not be averaged or interpreted as a performance trend because their sample units, populations, and settings differ. CNF page 44 also differs from parts of the paper, making source and result-version identification essential.

Source: 2024 paper, pp. 129–131; Rayong abstract, pp. 1–3; CNF 2026, p. 44

05

5. Distinguish publications, slides, and plans

The 2024 work has an official article record. Rayong is available as an abstract and slides with unverified peer review. Demonstrations show presented functionality, the GPO Ignite 2025 award records an innovation milestone, and the at-least-five-hospital target is a plan. Each answers a different kind of question.

SaMD documentation work and references to standards do not establish authorization. Agreement with clinicians in one dataset does not prove reduced mortality or replacement of standard examination. Each claim requires evidence addressing that specific outcome.

Source: 2024 paper, pp. 123–132; CNF 2026, pp. 42, 52–60; GPO 2026, pp. 8, 11; EasyEF project brief, 22 June 2026

06

6. Questions for evaluating collaboration

Begin by asking who was studied, which clips were used, what was excluded, how the reference was established, and which system version was evaluated. Then ask whether every category is reported, especially the intermediate group, and how unusable data was handled.

Finally, separate classification results from service outcomes such as task duration, workload, and referral patterns. The project’s planned User Acceptance and Deployment Feasibility evaluations address this additional layer. Stating what remains unknown helps define a clearer collaboration.

Source: 2024 paper, pp. 128–132; Rayong abstract, pp. 1–3; EasyEF project brief, 22 June 2026

SOURCE NOTES

Sources for this page

  1. บทความวิจัยปี 2567 / 2024 original article — J Prapokklao Hosp Clin Med Educat Center 41(2):123–132 ↗
  2. บทคัดย่อการศึกษาระยอง ไทย–อังกฤษ หน้า 1–3 / Rayong study abstract, Thai and English, pp. 1–3; publication and peer-review status not verified / ยังไม่ยืนยันสถานะตีพิมพ์และการทบทวนโดยผู้ทรงคุณวุฒิ
  3. เอกสารนำเสนอ / Presentation: EasyEF Lunch Symposium, CNF 17th, 26 March 2026 — หน้า / pp. 24, 42, 44, 52–60
  4. เอกสารนำเสนอ / Presentation: EasyEF AI Echo, GPO, 4 February 2026 — หน้า / pp. 8, 11
  5. ข้อเสนอโครงการ EasyEF ฉบับ 22 มิถุนายน 2569: ขอบเขตและแผนงาน / EasyEF project brief, 22 June 2026: scope and planned work

Presentations reflect information at the time they were prepared. Explore context and evidence status in the resource library.

Explore the source library →