extraction idea: process article tables through llm for clean up

[adelavega]

testing on chatgpt, it does a good job of splitting up tables into sub tables corresponding to a single analyses, regardless of idiosyncractic arrangments in a journal.

this could be useful in parsing complex tables into individual analyses, which can later be further probed for contrast specific information / annotations / semanatic information

key will be to develop a heuristic for when to apply this or ensure it happens ina cost effect manner (i.e. without passing in the whole text)

e.g.: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7913329/

Table 4: Regression analysis with SP and SR scores

Men + Women: Positive correlation: Dollar Win > Nil vs. SR score

Cluster Size (k)	Voxel (peak Z)	x	y	z	Side	Brain Region
334	4.66	-30	17	43	L	Middle frontal gyrus
819	4.51	36	-28	49	R	Postcentral gyrus
421	4.23	27	-40	1	R	Lingual gyrus
180	4.19	6	-52	28	L/R	PCC/PCu

Men + Women: Positive correlation: Cent Loss > Nil vs. SR score

Cluster Size (k)	Voxel (peak Z)	x	y	z	Side	Brain Region
121	4.52	60	-46	43	R	Inferior parietal lobule
162	4.29	-3	26	49	L/R	Dorsomedial prefrontal cortex
104	4.03	33	26	43	R	Middle frontal gyrus

Men + Women: Negative correlation: Dollar Win > Nil vs. SP score

Cluster Size (k)	Voxel (peak Z)	x	y	z	Side	Brain Region
222	4.50	42	41	13	R	Middle frontal gyrus

Men: Positive correlation: Dollar Win > Nil vs. SR score

Cluster Size (k)	Voxel (peak Z)	x	y	z	Side	Brain Region
180	4.19	6	-52	28	L/R	PCC/PCu

Women: Negative correlation: Dollar Win > Nil vs. SP score

Cluster Size (k)	Voxel (peak Z)	x	y	z	Side	Brain Region
268	4.28	45	38	13	R	Middle frontal gyrus
124	4.16	51	-22	43	R	Postcentral gyrus

Women: Negative correlation: Dollar Win > Cent Win vs. SP score

Cluster Size (k)	Voxel (peak Z)	x	y	z	Side	Brain Region
105	4.75	36	26	-11	R	Anterior insula
108	4.24	-15	38	43	L	Superior frontal gyrus
71	4.22	54	5	-29	R	Temporal gyrus