FEATURE 374

/ Medical Imaging and Oncology

This page shows the detailed analysis of a specific feature in the Sparse Autoencoder (SAE). It includes the semantic interpretation generated by the LLM, the top activating documents that trigger this feature, and statistical metrics like density and activation distribution.

Semantic Interpretation

gemma3:12b #8

The neuron strongly activates in contexts related to medical imaging (CT, MRI), particularly in the diagnosis and treatment of cancers. This includes descriptions of imaging techniques, tumor characteristics, staging, and treatment strategies. The focus is on specific cancers (pancreatic, breast, gastric, laryngeal, intracranial) and often involves discussions of surgical resectability, vascular involvement, and treatment response. The presence of terms like 'adenocarcinoma', 'carcinoma', 'metastases', 'tumor', 'angiogenesis', 'resectable', 'CT', 'MRI', 'lymph node', and descriptions of imaging findings (e.g., 'dilated gastrocolic trunk', 'bone destruction') are key indicators. The negative examples, dealing with topics like protein sampling, teat dips, and obsessive-compulsive disorder, lack this medical imaging and oncology focus.

STATISTICS & DISTRIBUTION

Density

0.02900

Peak Act

3.72

0.0 Max

Global Context

TOP ACTIVATING CONTEXTS

DOC #300 ANALYZE

ACT: 3.7245

Adenocarcinoma of the head of the pancreas: determination of surgical unresectability with thin-section pancreatic-phase helical CT. This study was conducted to evaluate newly introduced criteria for unresectability of pancreatic cancer with thin-section pancreatic-phase helical CT. Twenty-five patients with adenocarcinoma in the head of the pancr…

DOC #550 ANALYZE

ACT: 2.9631

Alterations in vascular gene expression in invasive breast carcinoma. The molecular signature that defines tumor microvasculature will likely provide clues as to how vascular-dependent tumor proliferation is regulated. Using purified endothelial cells, we generated a database of gene expression changes accompanying vascular proliferation in invasi…

DOC #94 ANALYZE

ACT: 2.9483

CT of intracranial metastases with skull and scalp involvement. Twenty-eight persons with contiguous intracranial skull, and often extracranial metastatic disease are reported. These lesions comprised 7.6% of a series of 250 consecutive patients with intracranial metastatic disease. Only three of 28 patients had other intracranial lesions and only…

DOC #54 ANALYZE

ACT: 2.6670

Pre-therapeutic evaluation of laryngeal carcinomas using computed tomography and magnetic resonance imaging. A prospective and comparative computed tomography (CT)/magnetic resonance imaging (MRI) study on 90 patients with endoscopically examined and histologically proven laryngeal malignancy is presented. Post-operative pathological and intra-rad…

DOC #289 ANALYZE

ACT: 2.5341

Skull Base Chondrosarcoma Caused by Ollier Disease: A Case Report and Literature Review. Ollier disease (OD) is a rare, nonhereditary bone disease that is characterized by the presence of multiple enchondromatosis (3 or more) with a typical asymmetric distribution which is mainly confined to the appendicular skeleton. OD's most serious complicatio…

TOPOLOGY

ROOT OF FAMILY #9

Medical Case Reports & Cancer Research #2232
Child of Metastatic Cancer
#1792

CORRELATIONS

W – Weight-space · similarity between decoder vectors (features that point in similar directions in the embedding space).

D – Data / co-activation · features that tend to fire together on the same documents (co-occurrence in the dataset).

Metastatic Cancer 0.34
Drug Resistance Mechanisms in Cancer 0.14
Medical Case Reports & Cancer Research 0.03
Feature 2596 0.03
Feature 954 0.03
Sensing and Detection Technologies 0.03

Ollama Model

Temp

K+

K-

System Prompt

You are a meticulous researcher investigating a specific neuron in a language model. Your task is to determine what behavior this neuron is responsible for: what concepts, topics, or linguistic features does it activate on?

INPUT DESCRIPTION: You will receive two inputs: 1) Maximum Activation Examples and 2) Zero Activation Examples.
1. You will be given several text examples that activate the neuron, along with a number indicating how strongly it was activated. This means there is some feature, concept, or pattern in this text that 'excites' this neuron.
2. You will also be given several text examples that do NOT activate the neuron. This means the feature or concept is not present in these texts.

OUTPUT DESCRIPTION: Given the inputs provided, complete the following tasks.
1. Based on the MAXIMUM ACTIVATION EXAMPLES, list potential topics, concepts, themes, and features they have in common. Be specific. You may need to look at different levels of granularity. List as many as possible. Give greater weight to concepts more prominent in higher-activation examples.
2. Based on the zero activation examples, systematically exclude any topic/concept/feature listed above that also appears in the zero activation examples.
3. Based on the two previous steps, perform a thorough analysis of which feature, concept, or topic, at which level of granularity, is likely to activate this neuron. Use Occam's razor, as long as it fits the evidence provided. Be highly rational and analytical.
4. Based on step 3, summarize this concept in 1-8 words, in the form FINAL: <explanation>. Do NOT return anything after these 1-8 words.

Respond EXCLUSIVELY with valid JSON: {'label': '...', 'description': '...'}

FEATURE 374

Semantic Interpretation

Statistics Explained

Global Context