MHSMA: The Modified Human Sperm Morphology Analysis Dataset

The MHSMA dataset is a collection of human sperm images from 235 patients with male factor infertility. Each image is labeled by experts for normal or abnormal sperm acrosome, head, vacuole, and tail.

The training, validation, and test sets contain 1000, 240, and 300 images, respectively.

Images are available in two different crop sizes: 128x128- and 64x64-pixel. The following figure shows two versions of the same instance.

128x128-pixel	64x64-pixel

In MHSMA, each instance is a grayscale image capturing a single sperm. The head of the sperm is roughly located at the center of the image. Also, the sperm tail is not entirely visible in the images.

Labels can be either 0 (normal, positive) or 1 (abnormal, negative).

The dataset is available in .npy format. You can load the .npy files using numpy.load. The details of the files are described in the table below.

File	Shape	Type	Description
x_128_train.npy	(1000, 128, 128)	uint8	Training set, 128x128-pixel version
x_128_valid.npy	(240, 128, 128)	uint8	Validation set, 128x128-pixel version
x_128_test.npy	(300, 128, 128)	uint8	Test set, 128x128-pixel version
x_64_train.npy	(1000, 64, 64)	uint8	Training set, 64x64-pixel version
x_64_valid.npy	(240, 64, 64)	uint8	Validation set, 64x64-pixel version
x_64_test.npy	(300, 64, 64)	uint8	Test set, 64x64-pixel version
y_acrosome_train.npy	(1000,)	uint8	Training set labels for acrosome
y_acrosome_valid.npy	(240,)	uint8	Validation set labels for acrosome
y_acrosome_test.npy	(300,)	uint8	Test set labels for acrosome
y_head_train.npy	(1000,)	uint8	Training set labels for head
y_head_valid.npy	(240,)	uint8	Validation set labels for head
y_head_test.npy	(300,)	uint8	Test set labels for head
y_vacuole_train.npy	(1000,)	uint8	Training set labels for vacuole
y_vacuole_valid.npy	(240,)	uint8	Validation set labels for vacuole
y_vacuole_test.npy	(300,)	uint8	Test set labels for vacuole
y_tail_train.npy	(1000,)	uint8	Training set labels for tail
y_tail_valid.npy	(240,)	uint8	Validation set labels for tail
y_tail_test.npy	(300,)	uint8	Test set labels for tail

The following table shows the number of positive and negative examples in the dataset.

Set	Label	# Positive	# Negative	% Positive
Whole dataset	Acrosome	1,086	454	70.52
	Head	1,122	418	72.86
	Vacuole	1,301	239	84.48
	Tail	1,471	69	95.52
Training set	Acrosome	699	301	69.90
	Head	727	273	72.70
	Vacuole	830	170	83.00
	Tail	954	46	95.40
Validation set	Acrosome	174	66	72.50
	Head	176	64	73.33
	Vacuole	209	31	87.08
	Tail	233	7	97.08
Test set	Acrosome	213	87	71.00
	Head	219	81	73.00
	Vacuole	262	38	87.33
	Tail	284	16	94.67

Results

If you would like to add a new result, you can open a pull request.

Method	Label	Accuracy	Precision	Recall	F0.5 score	G-mean	ROC AUC	MCC
A novel deep learning method for automatic assessment of human sperm images (Apr 2019)	Acrosome	76.67	85.93	80.28	84.74	83.06	83.89	+0.4618
	Head	77.00	83.48	85.39	83.86	84.43	77.80	+0.4053
	Vacuole	91.33	94.36	95.80	94.65	95.08	88.08	+0.5910
Effect of Deep Transfer and Multi-task Learning on Sperm Abnormality Detection (Nov 2020)	Acrosome (DTL)	79.00	80.24	93.42	82.57	86.58	79.65	+0.4447
	Acrosome (DMTL)	80.66	82.42	92.48	84.26	87.31	78.19	+0.4984
	Head (DTL)	84.00	87.01	91.78	87.92	89.36	81.56	+0.5775
	Head (DMTL)	82.00	82.60	95.43	84.89	88.78	78.40	+0.5021
	Vacuole (DTL)	94.00	95.18	98.09	95.75	96.62	94.73	+0.7082
	Vacuole (DMTL)	92.33	94.75	96.56	95.11	95.65	93.64	+0.6348

Citation

If you use this dataset in your research, please kindly cite our work as:

@article{javadi2019novel,
  title={A novel deep learning method for automatic assessment of human sperm images},
  author={Javadi, Soroush and Mirroshandel, Seyed Abolghasem},
  journal={Computers in Biology and Medicine},
  volume={109},
  pages={182--194},
  year={2019},
  doi={10.1016/j.compbiomed.2019.04.030}
}

License

This dataset is made available under the CC BY-NC-SA 4.0 license.

Credits

MHSMA is based on the Human Sperm Morphology Analysis Dataset (HSMA-DS) (Ghasemian et al., 2015).