Histopathological Transfer Learning for Acute Lymphoblastic Leukemia Detection
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The detection of Acute Lymphoblastic (or Lymphocytic) Leukemia (ALL) is being increasingly performed with the help of Computer Aided Diagnosis (CAD) systems based on Deep Learning (DL), that support the pathologists in performing their decision by analyzing the blood samples to determine the presence of lymphoblasts. When using DL, the limited dimensionality of ALL databases favors the use of transfer learning techniques to increase the accuracy in the detection, by considering Convolutional Neural Networks (CNN) pretrained on the general purpose ImageNet database. However, no method in the literature has yet considered the use of CNNs pretrained on histopathology databases to perform transfer learning for ALL detection. In fact, the majority of histopathology databases in the literature has either a small number of samples or limited ground truth labeling possibilities (e.g., only two possible classes), which hinders the effectiveness of training CNNs from scratch. In this paper, we propose the first method based on histopathological transfer learning for ALL detection, that trains a CNN on a histopathology database to classify tissue types, then performs a fine tuning on the ALL database to detect the presence of lymphoblasts. As histopathology database, we consider a multi-label dataset with a significantly higher number of samples and classes with respect to the literature, which enables CNNs to learn general features for histopathology image processing and hence allow to perform a more effective transfer learning, with respect to CNNs pretrained on ImageNet. We evaluate the methodology on a publicly-available ALL database and considering multiple CNNs, with results confirming the validity of our approach.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 it