Abstract B030: Multiple instance learning of large-scale DNA organization to characterize prostate cancer aggressiveness
Bibliographic record
Abstract
Abstract Multiple instance learning (MIL) has become a popular approach to analyze histopathology datasets due to its weakly supervised nature. In particular, attention-based models can learn key instances of which labels are usually unknown or unavailable. For instance, large-scale DNA organization (LDO) analysis aims to make patient prognoses from quantitative features of nuclear morphometry and chromatin condensation calculated from images of the nucleus. Leveraging attention mechanisms allows a model to identify nuclei with alterations which likely contribute to patient outcome without individual cell labels. However, a crucial assumption of MIL that is often overlooked in histopathology applications, wherein a bag is positive if it has at least one positive instance. In a cancer context, this assumption is not robust, as a single malignantly transformed cell may be necessary but insufficient to cause carcinogenesis or malignant cancer progression. A reasonable adjustment is to learn a tolerable threshold of aberrant cells. The attention mechanism can be modified so that both key aggressive and indolent nuclei are identified in contrast to traditional MIL attention mechanisms which only give weight to positive instances. We demonstrate that this is an effective approach for prostate cancer (PCa) prognosis. A binary MIL classifier was trained to identify PCa patients (bags) of the indolent (negative) and aggressive (positive) outcomes. The cohort includes 38 Gleason score (GS) 6 patients who did not display signs of progression during active surveillance (AS) and 22 patients with GS 9 who died within 2 years of consultation. A linear attention layer identifies aggressive and indolent nuclei (instances) to generate a weighted mean representation of LDO features for the patient, reducing the influence of nuclei of ambiguous labels. To mitigate overfitting, weights are shared between the attention layer and patient classification layer and trained to optimize a combination of binary cross entropy loss on the nuclear and patient level. Patients in the training set were classified with a balanced accuracy of 0.851, and an F1-score of 0.815. This performance also translated to the patients in the holdout set, where the balanced accuracy and F1-scores of patient classification was 0.857 and 0.833 respectively. Tests on an independent cohort of 147 patients with GS 7+ also demonstrate that LDO score is correlated with GS, biochemical recurrence following brachytherapy, and progression in GS6 active surveillance patients. Future studies will analyze the performance of the trained classifiers on patients with GS7 and GS8 further, such as the classifier’s ability to rank severity of clinical outcomes with c-index. These results demonstrate the potential of the MIL-based LDO biomarker for prostate cancer patient prognosis and management. Further validation on the brachytherapy-treated cohort, and survival analysis will be done to assess the performance of the MIL classifier, and its potential benefit for prostate cancer management. Citation Format: Fumiya Inaba, Zhaoyang Chen, Anita Carraro, Paul Gallagher, Mira Keyes, Martial Guillaud, Calum MacAulay. Multiple instance learning of large-scale DNA organization to characterize prostate cancer aggressiveness [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr B030.
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 | 0.002 | 0.004 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 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".