MedIMeta: A comprehensive and easy-to-use multi-domain multi-task medical imaging meta-dataset

<p>We introduce the Medical Imaging Meta-Dataset (MedIMeta), a novel multi-domain, multi-task meta-dataset designed to facilitate the development and standardised evaluation of ML models and cross-domain few-shot learning algorithms for medical image classification. MedIMeta contains 19 medical imaging datasets spanning 10 different domains and encompassing 54 distinct medical tasks, offering opportunities for both single-task and multi-task training. All tasks are standardised to the same format and readily usable in PyTorch or other ML frameworks. All datasets have been previously published with an open license that allows&nbsp;redistribution or we obtained an explicit permission to do so.</p> <p>Each dataset within the MedIMeta dataset is standardized to a size of 224 &times; 224 pixels which matches image size commonly used in pre-trained models. Furthermore, the dataset comes with pre-made splits to ensure ease of use and standardized benchmarking. We release a user-friendly Python package to directly load images for use in PyTorch.<br><br></p> <h3>Links</h3> <ul> <li>Project website: <a href="https://www.woerner.eu/projects/medimeta/" target="_blank" rel="noopener">https://www.woerner.eu/projects/medimeta/</a></li> <li>Data loading code (medimeta Python package):&nbsp;<a href="https://github.com/StefanoWoerner/medimeta-pytorch" target="_blank" rel="noopener">https://github.com/StefanoWoerner/medimeta-pytorch</a></li> <li>Data creation code: <a href="https://github.com/StefanoWoerner/medimeta-dataset-scripts" target="_blank" rel="noopener">https://github.com/StefanoWoerner/medimeta-dataset-scripts</a></li> </ul> <p>&nbsp;</p> <h3>Dataset Overview</h3> <table> <tbody> <tr> <td><strong>Dataset Name</strong></td> <td><strong>Dataset ID</strong></td> <td><strong>License</strong></td> <td><strong>Domain</strong></td> <td><strong>Task Names</strong></td> <td><strong>Task Targets</strong></td> <td><strong># Labels</strong></td> </tr> <tr> <td>AML Cytomorphology</td> <td>aml</td> <td>CC BY-SA 4.0</td> <td>Microscopy</td> <td>morphological class</td> <td>multi-class classification</td> <td>15</td> </tr> <tr> <td>Breast Ultrasound</td> <td>bus</td> <td>CC BY-SA 4.0</td> <td>Breast ultrasound</td> <td>case category<br>malignancy</td> <td>multi-class classification<br>binary classification</td> <td>3<br>2</td> </tr> <tr> <td>Colorectal Cancer Histopathology</td> <td>crc</td> <td>CC BY-SA 4.0</td> <td>Histopathology</td> <td>tissue class</td> <td>multi-class classification</td> <td>9</td> </tr> <tr> <td>Chest X-ray Multi-disease</td> <td>cxr</td> <td>CC BY-SA 4.0</td> <td>Chest X-ray</td> <td>disease labels<br>patient sex</td> <td>multi-label classification<br>binary classification</td> <td>14<br>2</td> </tr> <tr> <td>Dermatoscopy</td> <td>derm</td> <td>CC BY-SA 4.0</td> <td>Dermatoscopy</td> <td>disease category</td> <td>multi-class classification</td> <td>7</td> </tr> <tr> <td>Diabetic Retinopathy (Regular Fundus)</td> <td>dr_regular</td> <td>CC BY-SA 4.0</td> <td>Retinal fundus</td> <td>DR level<br>Overall quality<br>Artifact<br>Clarity<br>Field definition</td> <td>ordinal regression<br>binary classification<br>ordinal regression<br>ordinal regression<br>ordinal regression</td> <td>5<br>2<br>6<br>5<br>5</td> </tr> <tr> <td>Diabetic Retinopathy (Ultra-widefield Fundus)</td> <td>dr_uwf</td> <td>CC BY-SA 4.0</td> <td>Retinal fundus</td> <td>DR level</td> <td>ordinal regression</td> <td>5</td> </tr> <tr> <td>Fundus Multi-disease</td> <td>fundus</td> <td>CC BY-SA 4.0</td> <td>Retinal fundus</td> <td>disease presence<br>disease labels</td> <td>binary classification<br>multi-label classification</td> <td>2<br>45</td> </tr> <tr> <td>Glaucoma-specific fundus images</td> <td>glaucoma</td> <td>CC BY-SA 4.0</td> <td>Retinal fundus</td> <td>Glaucoma suspect</td> <td>binary classification</td> <td>2</td> </tr> <tr> <td>Mammography (Calcifications)</td> <td>mammo_calc</td> <td>CC BY-SA 4.0</td> <td>Mammography</td> <td>pathology<br>calc type<br>calc distribution</td> <td>binary classification<br>multi-label classification<br>multi-label classification</td> <td>2<br>14<br>5</td> </tr> <tr> <td>Mammography (Masses)</td> <td>mammo_mass</td> <td>CC BY-SA 4.0</td> <td>Mammography</td> <td>pathology<br>mass shape<br>mass margins</td> <td>binary classification<br>multi-label classification<br>multi-label classification</td> <td>2<br>8<br>5</td> </tr> <tr> <td>OCT</td> <td>oct</td> <td>CC BY-SA 4.0</td> <td>OCT</td> <td>disease class<br>urgent referral</td> <td>multi-class classification<br>binary classification</td> <td>4<br>2</td> </tr> <tr> <td>Axial Organ Slices</td> <td>organs_axial</td> <td>CC BY-NC-SA 4.0</td> <td>Abdominal CT</td> <td>organ label</td> <td>multi-class classification</td> <td>11</td> </tr> <tr> <td>Coronal Organ Slices</td> <td>organs_coronal</td> <td>CC BY-NC-SA 4.0</td> <td>Abdominal CT</td> <td>organ label</td> <td>multi-class classification</td> <td>11</td> </tr> <tr> <td>Sagittal Organ Slices</td> <td>organs_sagittal</td> <td>CC BY-NC-SA 4.0</td> <td>Abdominal CT</td> <td>organ label</td> <td>multi-class classification</td> <td>11</td> </tr> <tr> <td>Peripheral Blood Cells</td> <td>pbc</td> <td>CC BY-SA 4.0</td> <td>Microscopy</td> <td>cell class</td> <td>multi-class classification</td> <td>8</td> </tr> <tr> <td>Pediatric Pneumonia</td> <td>pneumonia</td> <td>CC BY-SA 4.0</td> <td>Chest X-ray</td> <td>pneumonia presence<br>disease class</td> <td>binary classification<br>multi-class classification</td> <td>2<br>3</td> </tr> <tr> <td>Skin Lesion Evaluation (Dermoscopy)</td> <td>skinl_derm</td> <td>CC BY-SA 4.0</td> <td>Dermatoscopy</td> <td>Diagnosis<br>Diagnosis grouped<br>Pigment Network<br>Blue Whitish Veil<br>Vascular Structures<br>Vascular Structures grouped<br>Pigmentation<br>Pigmentation grouped<br>Streaks<br>Dots and Globules<br>Regression Structures<br>Regression Structures grouped</td> <td>multi-class classification<br>multi-class classification<br>multi-class classification<br>binary classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>binary classification</td> <td>15<br>5<br>3<br>2<br>8<br>3<br>5<br>3<br>3<br>3<br>4<br>2</td> </tr> <tr> <td>Skin Lesion Evaluation (Clinical Photography)</td> <td>skinl_photo</td> <td>CC BY-SA 4.0</td> <td>Clinical skin imaging</td> <td>Diagnosis<br>Diagnosis grouped<br>Pigment Network<br>Blue Whitish Veil<br>Vascular Structures<br>Vascular Structures grouped<br>Pigmentation<br>Pigmentation grouped<br>Streaks<br>Dots and Globules<br>Regression Structures<br>Regression Structures grouped</td> <td>multi-class classification<br>multi-class classification<br>multi-class classification<br>binary classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>multi-class classification<br>binary classification</td> <td>15<br>5<br>3<br>2<br>8<br>3<br>5<br>3<br>3<br>3<br>4<br>2</td> </tr> </tbody> </table>