sebasmos commited on
Commit
9ff479f
Β·
1 Parent(s): 8955ba4

updating progress medgemma results

Browse files
README.md CHANGED
@@ -15,7 +15,7 @@ datasets:
15
 
16
  Pretrained checkpoints for **"The Unawareness of AI Looking for Health Insurance Type from Normal Chest X-ray Images"**.
17
 
18
- All models were trained on [MIMIC-CXR-JPG v2.0.0](https://physionet.org/content/mimic-cxr-jpg/2.0.0/) with seed=123.
19
 
20
  ## Models
21
 
@@ -83,7 +83,33 @@ insurance_paper_weights/
83
  β”œβ”€β”€ exp4-1/ # Frequency filtering β€” Swin Transformer
84
  β”‚ β”œβ”€β”€ highpass/ └── lowpass/
85
  β”‚
86
- └── exp4-2/ # Frequency filtering β€” DenseNet
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
87
  β”œβ”€β”€ highpass/ └── lowpass/
88
  ```
89
 
@@ -107,7 +133,23 @@ insurance_paper_weights/
107
  | **exp4-1** | Freq filtering (HP+LP) | SwinT | 8 freq x 2 | 16 | `exp4-1/{highpass,lowpass}/{F}Hz.pt` | Yes |
108
  | **exp4-2** | Freq filtering (HP+LP) | DenseNet | 8 freq x 2 | 16 | `exp4-2/{highpass,lowpass}/{F}Hz.pt` | Yes |
109
 
110
- **Total: 153 unique checkpoints**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
111
 
112
  - **Demo combos** (exp3): `sex`, `age`, `race`, `sexage`, `sexrace`, `agerace`, `sexagerace`
113
  - **Frequencies** (exp4): 1, 5, 10, 25, 50, 100, 200, 400 Hz
@@ -176,10 +218,17 @@ Weights in JAMA_codes were trained on ORCD. Weights marked with (*) were trained
176
  | `exp4/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes/{F}{HighPass,LowPass}_weights/mamba/{direction}/Rand123/...` |
177
  | `exp4-1/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes/{F}{HighPass,LowPass}_swin_freq/Rand123/...` |
178
  | `exp4-2/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes/{F}{HighPass,LowPass}_densenet_freq/Rand123/...` |
 
 
 
 
 
 
 
179
 
180
  ## Bootstrap Evaluation Status
181
 
182
- All 153 bootstrap evaluations complete (n_bootstrap=20, sample_size=1000, seed=42).
183
 
184
  | Experiment | Evaluations | Status |
185
  |-----------|-------------|--------|
@@ -191,11 +240,20 @@ All 153 bootstrap evaluations complete (n_bootstrap=20, sample_size=1000, seed=4
191
  | **exp3/3-1/3-2** | 21 (3 models x 7 combos) | Done |
192
  | **exp4/4-1/4-2** | 48 (3 models x 2 directions x 8 freqs) | Done |
193
 
194
- Results: `bootstrap_results/` in the [code repository](https://github.com/TODO).
 
 
 
 
 
 
 
 
195
 
196
  ## Training Details
197
 
198
- - **Dataset**: MIMIC-CXR-JPG v2.0.0 (normal frontal chest X-rays)
 
199
  - **Task**: Binary classification β€” Private vs. Public/Government insurance
200
  - **Image size**: 448 x 448
201
  - **Seed**: 123
@@ -213,4 +271,4 @@ Results: `bootstrap_results/` in the [code repository](https://github.com/TODO).
213
 
214
  ## Code
215
 
216
- [Code Repository](https://github.com/TODO) <!-- Update with actual repo URL -->
 
15
 
16
  Pretrained checkpoints for **"The Unawareness of AI Looking for Health Insurance Type from Normal Chest X-ray Images"**.
17
 
18
+ All models were trained on [MIMIC-CXR-JPG v2.0.0](https://physionet.org/content/mimic-cxr-jpg/2.0.0/) with seed=123. MedGemma experiments use the MedGemma-refined subset (`mimic-cxr-gemma`), containing only normal CXRs filtered by MedGemma.
19
 
20
  ## Models
21
 
 
83
  β”œβ”€β”€ exp4-1/ # Frequency filtering β€” Swin Transformer
84
  β”‚ β”œβ”€β”€ highpass/ └── lowpass/
85
  β”‚
86
+ β”œβ”€β”€ exp4-2/ # Frequency filtering β€” DenseNet
87
+ β”‚ β”œβ”€β”€ highpass/ └── lowpass/
88
+ β”‚
89
+ β”œβ”€β”€ exp2_medgemma/ # MedGemma Resolution β€” MedMamba
90
+ β”‚ └── mamba_2.pt ... mamba_224.pt (8 files)
91
+ β”‚
92
+ β”œβ”€β”€ exp3_medgemma/ # MedGemma Demographics β€” MedMamba
93
+ β”‚ β”œβ”€β”€ sex.pt, age.pt, race.pt
94
+ β”‚ β”œβ”€β”€ sexage.pt, sexrace.pt, agerace.pt
95
+ β”‚ └── sexagerace.pt
96
+ β”‚
97
+ β”œβ”€β”€ exp3-1_medgemma/ # MedGemma Demographics β€” DenseNet
98
+ β”‚ └── (same 7 files)
99
+ β”‚
100
+ β”œβ”€β”€ exp3-2_medgemma/ # MedGemma Demographics β€” Swin Transformer
101
+ β”‚ └── (same 7 files)
102
+ β”‚
103
+ β”œβ”€β”€ exp4_medgemma/ # MedGemma Frequency filtering β€” MedMamba
104
+ β”‚ β”œβ”€β”€ highpass/
105
+ β”‚ β”‚ └── 1Hz.pt, 5Hz.pt, ... 400Hz.pt
106
+ β”‚ └── lowpass/
107
+ β”‚ └── 1Hz.pt, 5Hz.pt, ... 400Hz.pt
108
+ β”‚
109
+ β”œβ”€β”€ exp4-1_medgemma/ # MedGemma Frequency filtering β€” Swin Transformer
110
+ β”‚ β”œβ”€β”€ highpass/ └── lowpass/
111
+ β”‚
112
+ └── exp4-2_medgemma/ # MedGemma Frequency filtering β€” DenseNet
113
  β”œβ”€β”€ highpass/ └── lowpass/
114
  ```
115
 
 
133
  | **exp4-1** | Freq filtering (HP+LP) | SwinT | 8 freq x 2 | 16 | `exp4-1/{highpass,lowpass}/{F}Hz.pt` | Yes |
134
  | **exp4-2** | Freq filtering (HP+LP) | DenseNet | 8 freq x 2 | 16 | `exp4-2/{highpass,lowpass}/{F}Hz.pt` | Yes |
135
 
136
+ **Original: 153 checkpoints**
137
+
138
+ ### MedGemma Experiments (mimic-cxr-gemma dataset)
139
+
140
+ | Experiment | Description | Model | Config | # Checkpoints | HF path | Available? |
141
+ |------------|-------------|-------|--------|---------------|---------|------------|
142
+ | **exp2** | Resolution | MedMamba | 8 resolutions | 8 | `exp2_medgemma/mamba_{N}.pt` | In progress |
143
+ | **exp3** | Demographics (addDemo) | MedMamba | 7 demo combos | 7 | `exp3_medgemma/{combo}.pt` | Yes |
144
+ | **exp3-1** | Demographics (addDemo) | DenseNet | 7 demo combos | 7 | `exp3-1_medgemma/{combo}.pt` | Yes |
145
+ | **exp3-2** | Demographics (addDemo) | SwinT | 7 demo combos | 7 | `exp3-2_medgemma/{combo}.pt` | Yes |
146
+ | **exp4** | Freq filtering (HP+LP) | MedMamba | 8 freq x 2 | 16 | `exp4_medgemma/{highpass,lowpass}/{F}Hz.pt` | Yes |
147
+ | **exp4-1** | Freq filtering (HP+LP) | SwinT | 8 freq x 2 | 16 | `exp4-1_medgemma/{highpass,lowpass}/{F}Hz.pt` | Yes |
148
+ | **exp4-2** | Freq filtering (HP+LP) | DenseNet | 8 freq x 2 | 16 | `exp4-2_medgemma/{highpass,lowpass}/{F}Hz.pt` | Yes |
149
+
150
+ **MedGemma: 77 checkpoints**
151
+
152
+ **Total: 153 original + 77 MedGemma = 230 unique checkpoints**
153
 
154
  - **Demo combos** (exp3): `sex`, `age`, `race`, `sexage`, `sexrace`, `agerace`, `sexagerace`
155
  - **Frequencies** (exp4): 1, 5, 10, 25, 50, 100, 200, 400 Hz
 
218
  | `exp4/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes/{F}{HighPass,LowPass}_weights/mamba/{direction}/Rand123/...` |
219
  | `exp4-1/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes/{F}{HighPass,LowPass}_swin_freq/Rand123/...` |
220
  | `exp4-2/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes/{F}{HighPass,LowPass}_densenet_freq/Rand123/...` |
221
+ | `exp2_medgemma/mamba_{N}.pt` | `JAMA_codes_medgemma/{N}_mg_exp2/mamba/Rand123/...` |
222
+ | `exp3_medgemma/{combo}.pt` | `JAMA_codes_medgemma/{combo}_mg_exp3/mamba/Rand123/...` |
223
+ | `exp3-1_medgemma/{combo}.pt` | `JAMA_codes_medgemma/{combo}_mg_exp3-1/densenet/Rand123/...` |
224
+ | `exp3-2_medgemma/{combo}.pt` | `JAMA_codes_medgemma/{combo}_mg_exp3-2/swinTF/Rand123/...` |
225
+ | `exp4_medgemma/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes_medgemma/{F}{HighPass,LowPass}_mg_exp4/mamba/Rand123/...` |
226
+ | `exp4-1_medgemma/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes_medgemma/{F}{HighPass,LowPass}_mg_exp4-1/swinTF/Rand123/...` |
227
+ | `exp4-2_medgemma/{highpass,lowpass}/{F}Hz.pt` | `JAMA_codes_medgemma/{F}{HighPass,LowPass}_mg_exp4-2/densenet/Rand123/...` |
228
 
229
  ## Bootstrap Evaluation Status
230
 
231
+ All 153 original bootstrap evaluations complete (n_bootstrap=20, sample_size=1000, seed=42).
232
 
233
  | Experiment | Evaluations | Status |
234
  |-----------|-------------|--------|
 
240
  | **exp3/3-1/3-2** | 21 (3 models x 7 combos) | Done |
241
  | **exp4/4-1/4-2** | 48 (3 models x 2 directions x 8 freqs) | Done |
242
 
243
+ ### MedGemma Bootstrap Status
244
+
245
+ | Experiment | Evaluations | Status |
246
+ |-----------|-------------|--------|
247
+ | **MedGemma exp2** | 8 (mamba x 8 resolutions) | In progress |
248
+ | **MedGemma exp3/3-1/3-2** | 21 (3 models x 7 combos) | Done |
249
+ | **MedGemma exp4/4-1/4-2** | 48 (3 models x 2 directions x 8 freqs) | Done |
250
+
251
+ Results: `bootstrap_results/` in the [code repository](https://github.com/altis5526/Insurance-Project-Journal-version-).
252
 
253
  ## Training Details
254
 
255
+ - **Dataset (original)**: MIMIC-CXR-JPG v2.0.0 (normal frontal chest X-rays)
256
+ - **Dataset (MedGemma)**: mimic-cxr-gemma (MedGemma-refined, normal CXRs only)
257
  - **Task**: Binary classification β€” Private vs. Public/Government insurance
258
  - **Image size**: 448 x 448
259
  - **Seed**: 123
 
271
 
272
  ## Code
273
 
274
+ [Code Repository](https://github.com/altis5526/Insurance-Project-Journal-version-) <!-- Update with actual repo URL -->
exp2_medgemma/mamba_112.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bbe2dc4ec391f3803fda409b900bfb1598be867d0737baf7a873ca4a65db236e
3
+ size 119575610
exp2_medgemma/mamba_224.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bc0bf78d3577f8672036c9bfd4b53eaed2020d85902ec41b2ca61ad46b41ef85
3
+ size 119575610
exp2_medgemma/mamba_56.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fb3c373cf188d8dcfab1aedfa6413f15da3b2e006e1fb671370b7b2ba9266e4b
3
+ size 119574978
exp3-2_medgemma/agerace.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a254795819feea156b278a3d6f757951ddb83eb65ee9e69aa88e36c634d91efd
3
+ size 228333454
exp3-2_medgemma/sexagerace.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d10ac3cf6a1eb60f46e24c15d966e6468c1d66adc6b9ecdbd90d8644274dc5a7
3
+ size 228334936
exp3-2_medgemma/sexrace.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:90eff869dc822102a1905e0a4a4d5853f401a0ad7feb3ca0d58ffe11459aef3b
3
+ size 228333454