Marc-HealthAI/Stellar-Class-5L-6-CNN

STELLAR-CLASS-5L-6 (Custom CNN) - Fetal Planes Classification

This model is a custom 5-block convolutional network, trained from scratch (no ImageNet pretraining) on single-channel 128x128 fetal ultrasound images, to classify fetal ultrasound standard planes.

Model Performance (Test Set)

• Accuracy: 0.9365
• Macro F1 Score: 0.9258
• Macro Precision: 0.9080
• Macro Recall: 0.9477

---

Training History

Training & Evaluation Metrics Summary

Overall Training Job Metrics

• Total Epochs Completed: 77.0
• Total Training Steps: 14,861
• Total Training Runtime: 6,722.42 seconds (~1.87 hours)
• Training Throughput: 114.84 samples/sec (2.871 steps/sec)
• Final Overall Training Loss: 0.3823

---

Epoch-by-Epoch Evaluation History

Epoch	Step	Eval Loss	Eval Accuracy	Precision (Macro)	Recall (Macro)	F1 Score (Macro)	Eval Runtime (s)	Eval Samples/s
1.0	193	1.0250	57.09%	0.6097	0.6750	0.5527	21.36	115.87
2.0	386	0.8398	66.34%	0.6641	0.7336	0.6458	21.28	116.32
3.0	579	0.8334	61.98%	0.7021	0.7044	0.6137	20.91	118.35
4.0	772	0.6603	75.76%	0.7305	0.8001	0.7357	20.75	119.30
5.0	965	0.6480	70.51%	0.7257	0.7715	0.6904	21.27	116.35
6.0	1158	0.5478	79.27%	0.7594	0.8348	0.7719	20.95	118.16
7.0	1351	0.5481	78.10%	0.7766	0.8285	0.7702	21.00	117.84
8.0	1544	0.4906	80.00%	0.7782	0.8557	0.7831	21.26	116.44
9.0	1737	0.4720	82.06%	0.7890	0.8522	0.8034	21.07	117.47
10.0	1930	0.4620	78.95%	0.7784	0.8561	0.7757	20.36	121.57
11.0	2123	0.4660	80.81%	0.8012	0.8428	0.8007	20.30	121.93
12.0	2316	0.3931	84.65%	0.8062	0.8802	0.8258	20.87	118.58
13.0	2509	0.4077	82.26%	0.7952	0.8759	0.8075	20.97	118.01
14.0	2702	0.3988	83.07%	0.8095	0.8802	0.8192	20.85	118.68
15.0	2895	0.3679	85.58%	0.8250	0.8875	0.8422	20.93	118.25
16.0	3088	0.4304	81.90%	0.8298	0.8477	0.8182	20.86	118.63
17.0	3281	0.3488	85.01%	0.8183	0.8890	0.8352	20.82	118.91
18.0	3474	0.3188	87.60%	0.8457	0.9020	0.8620	20.99	117.94
19.0	3667	0.3132	87.27%	0.8423	0.8997	0.8595	20.35	121.61
20.0	3860	0.2414	95.31%	0.9391	0.9539	0.9461	28.11	88.06
21.0	4053	0.2907	88.57%	0.8499	0.9097	0.8725	19.90	124.40
22.0	4246	0.2894	88.44%	0.8554	0.9101	0.8745	19.96	124.02
23.0	4439	0.2929	87.23%	0.8343	0.9082	0.8549	19.67	125.84
24.0	4632	0.2975	87.47%	0.8522	0.9028	0.8672	19.20	128.91
25.0	4825	0.2632	89.78%	0.8680	0.9194	0.8876	19.58	126.41
26.0	5018	0.3043	86.14%	0.8451	0.9049	0.8594	20.02	123.62
27.0	5211	0.2777	86.99%	0.8392	0.9097	0.8580	20.95	118.14
28.0	5404	0.2746	87.35%	0.8413	0.9146	0.8632	20.85	118.69
29.0	5597	0.2574	89.05%	0.8559	0.9208	0.8772	21.34	115.98
30.0	5790	0.2353	89.78%	0.8667	0.9274	0.8886	21.49	115.20
31.0	5983	0.2704	87.64%	0.8691	0.9122	0.8786	20.84	118.75
32.0	6176	0.2784	87.31%	0.8637	0.9108	0.8746	19.96	124.01
33.0	6369	0.2325	90.18%	0.8745	0.9276	0.8944	20.41	121.29
34.0	6562	0.2369	88.73%	0.8629	0.9223	0.8829	20.15	122.84
35.0	6755	0.2102	91.88%	0.8942	0.9346	0.9109	20.03	123.58
36.0	6948	0.2328	90.34%	0.8671	0.9317	0.8902	19.20	128.89
37.0	7141	0.2320	89.62%	0.8669	0.9292	0.8889	19.50	126.94
38.0	7334	0.2381	89.33%	0.8726	0.9244	0.8899	19.55	126.60
39.0	7527	0.2251	90.79%	0.8845	0.9298	0.9014	19.50	126.89
40.0	7720	0.2074	92.00%	0.8963	0.9373	0.9128	19.47	127.09
41.0	7913	0.1962	92.40%	0.8989	0.9390	0.9161	19.22	128.74
42.0	8106	0.2235	89.86%	0.8798	0.9279	0.8965	19.05	129.92
43.0	8299	0.2132	91.15%	0.8893	0.9341	0.9064	19.67	125.83
44.0	8492	0.1945	92.20%	0.8913	0.9418	0.9120	20.48	120.87
45.0	8685	0.2051	91.47%	0.8945	0.9383	0.9110	20.70	119.56
46.0	8878	0.1984	91.72%	0.8907	0.9396	0.9102	21.08	117.40
47.0	9071	0.2043	90.99%	0.8812	0.9390	0.9030	20.95	118.14
48.0	9264	0.1873	92.20%	0.9008	0.9392	0.9169	20.68	119.70
49.0	9457	0.1938	91.60%	0.8969	0.9393	0.9127	19.54	126.66
50.0	9650	0.1880	92.20%	0.8996	0.9418	0.9168	19.51	126.88
51.0	9843	0.2019	91.27%	0.8935	0.9376	0.9105	20.00	123.73
52.0	10036	0.1912	92.69%	0.9084	0.9407	0.9222	22.24	111.26
53.0	10229	0.1808	92.77%	0.9017	0.9437	0.9195	19.82	124.85
54.0	10422	0.1802	92.73%	0.9060	0.9443	0.9220	20.44	121.08
55.0	10615	0.1847	92.24%	0.8988	0.9413	0.9160	22.05	112.24
56.0	10808	0.1777	93.05%	0.9089	0.9450	0.9244	20.32	121.80
57.0	11001	0.1736	92.77%	0.9045	0.9431	0.9212	19.59	126.33
58.0	11194	0.1752	92.20%	0.8952	0.9456	0.9155	19.27	128.45
59.0	11387	0.1764	92.57%	0.8991	0.9470	0.9188	19.45	127.22
60.0	11580	0.1735	93.05%	0.9078	0.9459	0.9241	20.13	122.95
61.0	11773	0.1684	92.69%	0.9004	0.9459	0.9194	21.63	114.41
62.0	11966	0.1734	92.97%	0.9065	0.9468	0.9235	20.99	117.91
63.0	12159	0.1681	93.45%	0.9102	0.9491	0.9272	18.93	130.73
64.0	12352	0.1695	93.74%	0.9158	0.9483	0.9304	19.99	123.80
65.0	12545	0.1768	92.53%	0.8996	0.9429	0.9180	19.23	128.71
66.0	12738	0.1658	93.21%	0.9111	0.9465	0.9264	19.83	124.79
67.0	12931	0.1648	93.74%	0.9178	0.9504	0.9322	21.18	116.88
68.0	13124	0.1677	93.58%	0.9143	0.9495	0.9298	21.31	116.13
69.0	13317	0.1640	93.29%	0.9079	0.9485	0.9254	19.90	124.37
70.0	13510	0.1677	93.29%	0.9082	0.9493	0.9259	19.83	124.83
71.0	13703	0.1685	92.69%	0.9047	0.9438	0.9211	19.43	127.39
72.0	13896	0.1725	92.40%	0.9016	0.9421	0.9184	19.38	127.72
73.0	14089	0.1603	94.14%	0.9216	0.9511	0.9349	21.37	115.84
74.0	14282	0.1608	93.86%	0.9162	0.9513	0.9317	21.07	117.45
75.0	14475	0.1687	92.85%	0.9072	0.9441	0.9228	19.52	126.83
76.0	14668	0.1655	93.13%	0.9045	0.9485	0.9231	19.71	125.60
77.0	14861	0.1652	93.21%	0.9111	0.9463	0.9263	21.09	117.33

---

Periodic Intermediate Training Log (Every 500 Steps)

Step	Epoch %	Loss	Gradient Norm	Learning Rate
500	2.59%	1.1721	3.5566	9.9835e-5
1000	5.18%	0.8184	3.2590	9.9340e-5
1500	7.77%	0.6885	4.2848	9.8519e-5
2000	10.36%	0.6012	5.8193	9.7376e-5
2500	12.95%	0.5511	5.3994	9.5920e-5
3000	15.54%	0.5048	3.5960	9.4160e-5
3500	18.13%	0.4711	4.5645	9.2107e-5
4000	20.73%	0.4305	4.2740	8.9776e-5
4500	23.32%	0.4042	3.5289	8.7181e-5
5000	25.91%	0.3907	3.2670	8.4340e-5
5500	28.50%	0.3668	6.7554	8.1272e-5
6000	31.09%	0.3539	7.4994	7.7996e-5
6500	33.68%	0.3327	2.6040	7.4536e-5
7000	36.27%	0.3124	2.1215	7.0913e-5
7500	38.86%	0.3119	6.6685	6.7151e-5
8000	41.45%	0.2941	8.4950	6.3276e-5
8500	44.04%	0.2845	4.3676	5.9313e-5
9000	46.63%	0.2782	2.6447	5.5288e-5
9500	49.22%	0.2787	7.9614	5.1229e-5
10000	51.81%	0.2562	2.3678	4.7161e-5
10500	54.40%	0.2575	3.7938	4.3112e-5
11000	56.99%	0.2502	2.5999	3.9109e-5
11500	59.59%	0.2423	7.0387	3.5177e-5
12000	62.18%	0.2328	3.6975	3.1344e-5
12500	64.77%	0.2356	4.8368	2.7635e-5
13000	67.36%	0.2237	7.0377	2.4073e-5
13500	69.95%	0.2230	5.3766	2.0683e-5
14000	72.54%	0.2236	4.5224	1.7487e-5
14500	75.13%	0.2161	10.0799	1.4507e-5

---

Key Highlights & Performance Analysis

1. Best Validation Loss:
   • Step 14,089 (Epoch 73.0): Achieved an evaluation loss of 0.1603 with an accuracy of 94.14% and Macro F1 of 0.9349.
2. Best Overall Accuracy & F1 Score:
   • Step 3,860 (Epoch 20.0): Reached a peak accuracy spike of 95.31% and Macro F1 of 0.9461.
   • Late Convergence Plateau: From Epoch 63 onwards, evaluation accuracy consistently held high around 93.0% – 94.1% with minimal loss fluctuations.
3. Training Loss Convergence:
   • Training loss decreased steadily from 1.1721 (Step 500) down to 0.2161 (Step 14,500), showing smooth convergence aligned with the learning rate schedule.

Dataset Information

This model was trained on dataset Marc-HealthAI/fetal-planes-classification-dataset-main.

• Classes: Fetal abdomen, Fetal brain, Fetal femur, Fetal thorax, Maternal cervix, Other
• Dataset Features: ['Image_name', 'Patient_num', 'Plane', 'Brain_plane', 'Operator', 'US_Machine', 'Split', 'image']

Original Dataset README Content

Click to expand original dataset README

Fetal_Planes_DB

Burgos-Artizzu, X.P., Coronado-Gutiérrez, D., Valenzuela-Alcaraz, B. et al. Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes. Sci Rep 10, 10200 (2020). https://doi.org/10.1038/s41598-020-67076-5

Data Description

A large dataset of routinely acquired maternal-fetal screening ultrasound images collected from two different hospitals by several operators and ultrasound machines. All images were manually labeled by an expert maternal fetal clinician (B.V-A.). Images were divided into 6 classes: four of the most widely used fetal anatomical planes (Abdomen, Brain, Femur and Thorax), the mother’s cervix (widely used for prematurity screening) and a general category to include any other less common image plane. Fetal brain images were further categorized into the 3 most common fetal brain planes (Trans-thalamic, Trans-cerebellum, Trans-ventricular) to judge fine grain categorization performance. The final dataset is comprised of over 12,400 images from 1,792 patients.

The dataset details are described in our open-acces paper: Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes

If you find this dataset useful, please cite:

@article{Burgos-ArtizzuFetalPlanesDataset,
  title={Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes},
  author={Burgos-Artizzu, X.P. and Coronado-Gutiérrez, D. and Valenzuela-Alcaraz, B. and Bonet-Carne, E. and Eixarch, E. and Crispi, F. and Gratacós, E.},
  journal={Nature Scientific Reports}, 
  volume={10},
  pages={10200},
  doi="10.1038/s41598-020-67076-5",
  year={2020}
}