Deep Neural Networks-based Malignant Breast Lesions Detection and Segmentation from Mammography
Bibliographic record
Abstract
Breast cancer has the highest incidence among all cancer types for women globally. Convolutional Neural Networks (CNN) is the most widely used models for computer-aided breast cancer detection in mammograms. This paper used CNN-based architecture for malignant lesion detection and segmentation in mammography images using the public CBIS-DDSM and an in-house (IMI) dataset. The CBIS-DDSM consists of 1000 images, and the IMI consists of 295 images. The CBIS-DDSM was randomly split into 80/10/10% of training/validation/ testing datasets. In this study, we applied the U-Net model and devised two setups: first, train and test the model with the CBIS-DDSM, and second, use transfer learning in which the pre-trained model on the CBIS-DDSM dataset is fine-tuned by the IMI train set and tested on the IMI test set. Evaluation metrics for calculating the similarity include the Dice coefficient, but we also reported the Intersection-Over-Union index. The Free-Response receiver operating characteristic curve (FROC) was chosen to evaluate the detection performance as an evaluation criterion. The dice and IOU values of the first setup were 0.98±0.006 and 0.97±003, respectively, which indicates that the predicted lesions are highly similar to ground-truth lesions. The FROC curve of setup1 shows that the model could detect lesions with the true positive rate of 0.99±0.17 at the false positives per image of 2.7. In setup 2, the pre-trained model with the CBISM-DDSM dataset tuned with the IMI achieved the dice and IOU of 0.92±0.0 and 0.98±0.008, respectively, on the IMI test set. Also, the model detected lesions with a true positive rate of 0.98±0.20 at the false positives per image of 3.3. Our study showed the potential of the Transfer learning techniques to guarantee the reliable performance of publicly developed models by tuning them with a low number of in-house patients.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".