使用PhysioNet epicmp评估AFib时FP计数不一致问题咨询

我正在使用PhysioNet的epicmp评估房颤（AFib）检测算法，发现一处疑似不一致的情况——可能是bug，也可能是我对预期行为理解有误。我采用长期房颤数据库中的04015记录进行测试。

参考标注（Referenz） #

[09:06:50.336]   102584 - [09:07:58.416]   119604 covers start of T3
[09:08:07.092]   121773 - [09:08:08.776]   122194 covers T3
[09:08:53.392]   133348 - [09:11:07.428]   166857 covers T4
[10:13:04.980]  1096245 - [10:13:12.216]  1098054 covered by T4
[10:15:41.184]  1135296 - [10:15:58.380]  1139595 x
[10:34:49.744]  1422436 - [10:34:54.192]  1423548 covered by Z8
[10:37:17.108]  1459277 - [10:37:21.664]  1460416 x

算法检测结果（Test） #

[09:02:36.720]    39180 - [09:02:52.876]    43219 x
[09:03:59.304]    59826 - [09:05:01.064]    75266 x
[09:06:59.724]   104931 - [09:08:13.108]   123277 overlaps end of R1, contains R2
[09:09:11.556]   137889 - [09:11:06.360]   166590 included in R3
[10:11:11.992]  1067998 - [10:13:30.488]  1102622 end overlaps start of R4
[10:19:10.596]  1187649 - [10:19:39.912]  1194978 x
[10:28:39.784]  1329946 - [10:29:04.792]  1336198 x
[10:33:49.716]  1407429 - [10:34:54.360]  1423590 covers R6

手动检测后，我识别出4个无对应参考片段的检测结果（应计为假阳性FP），但epicmp报告FP=3，二者计数不符。请问这是epicmp的预期行为吗？是否存在匹配规则、容差窗口或边缘情况处理导致该差异？我将尝试附上自定义AFib标注文件以供排查，感谢您的解答！

内容的提问来源于stack exchange，提问作者ralfiii